Search results for: ensemble classifier by voting

Authors: N. Harun, E. E. Masiren, W. H. W. Ibrahim, F. Adam

Abstract:

Amine absorption process is an approach for mitigation of CO2 from flue gas that produces from power plant. This process is the most common system used in chemical and oil industries for gas purification to remove acid gases. On the challenges of this process is high energy requirement for solvent regeneration to release CO2. In the past few years, mixed alkanolamines have received increasing attention. In most cases, the mixtures contain N-methyldiethanolamine (MDEA) as the base amine with the addition of one or two more reactive amines such as PZ. The reason for the application of such blend amine is to take advantage of high reaction rate of CO2 with the activator combined with the advantages of the low heat of regeneration of MDEA. Several experimental and simulation studies have been undertaken to understand this process using blend MDEA/PZ solvent. Despite those studies, the mechanism of CO2 absorption into the aqueous MDEA is not well understood and available knowledge within the open literature is limited. The aim of this study is to investigate the intermolecular interaction of the blend MDEA/PZ using Molecular Dynamics (MD) simulation. MD simulation was run under condition 313K and 1 atm using NVE ensemble at 200ps and NVT ensemble at 1ns. The results were interpreted in term of Radial Distribution Function (RDF) analysis through two system of interest i.e binary and tertiary. The binary system will explain the interaction between amine and water molecule while tertiary system used to determine the interaction between the amine and CO2 molecule. For the binary system, it was observed that the –OH group of MDEA is more attracted to water molecule compared to –NH group of MDEA. The –OH group of MDEA can form the hydrogen bond with water that will assist the solubility of MDEA in water. The intermolecular interaction probability of –OH and –NH group of MDEA with CO2 in blended MDEA/PZ is higher than using single MDEA. This findings show that PZ molecule act as an activator to promote the intermolecular interaction between MDEA and CO2.Thus, blend of MDEA with PZ is expecting to increase the absorption rate of CO2 and reduce the heat regeneration requirement.

Keywords: amine absorption process, blend MDEA/PZ, CO2 capture, molecular dynamic simulation, radial distribution function

Procedia PDF Downloads 295

Authors: Kemka Ihemelandu, Chukwuemeka Ihemelandu

Abstract:

Melanoma remains a public health crisis, with incidence rates increasing rapidly in the past decades. Improving diagnostic accuracy to decrease misdiagnosis using Artificial intelligence (AI) continues to be documented. Unfortunately, unintended racially biased outcomes, a product of lack of diversity in the dataset used, with a noted class imbalance favoring lighter vs. darker skin tone, have increasingly been recognized as a problem.Resulting in noted limitations of the accuracy of the Convolutional neural network (CNN)models. CNN models are prone to biased output due to biases in the dataset used to train them. Our aim in this study was the optimization of convolutional neural network algorithms to mitigate bias in the automated diagnosis of melanoma. We hypothesized that our proposed training algorithms based on a data augmentation method to optimize the diagnostic accuracy of a CNN classifier by generating new training samples from the original ones will reduce bias in the automated diagnosis of melanoma. We applied geometric transformation, including; rotations, translations, scale change, flipping, and shearing. Resulting in a CNN model that provided a modifiedinput data making for a model that could learn subtle racial features. Optimal selection of the momentum and batch hyperparameter increased our model accuracy. We show that our augmented model reduces bias while maintaining accuracy in the automated diagnosis of melanoma.

Keywords: bias, augmentation, melanoma, convolutional neural network

Procedia PDF Downloads 211

Authors: Chung-Ming Lo, Chung-Chien Lee

Abstract:

In many countries, the lifetime prevalence of shoulder pain is up to 70%. In America, the health care system spends 7 billion per year about the healthy issues of shoulder pain. With respect to the origin, up to 70% of shoulder pain is attributed to rotator cuff lesions This study proposed a computer-aided diagnosis (CAD) system to assist radiologists classifying rotator cuff lesions with less operator dependence. Quantitative features were extracted from the shoulder ultrasound images acquired using an ALOKA alpha-6 US scanner (Hitachi-Aloka Medical, Tokyo, Japan) with linear array probe (scan width: 36mm) ranging from 5 to 13 MHz. During examination, the postures of the examined patients are standard sitting position and are followed by the regular routine. After acquisition, the shoulder US images were drawn out from the scanner and stored as 8-bit images with pixel value ranging from 0 to 255. Upon the sonographic appearance, the boundary of each lesion was delineated by a physician to indicate the specific pattern for analysis. The three lesion categories for classification were composed of 20 cases of tendon inflammation, 18 cases of calcific tendonitis, and 18 cases of supraspinatus tear. For each lesion, second-order statistics were quantified in the feature extraction. The second-order statistics were the texture features describing the correlations between adjacent pixels in a lesion. Because echogenicity patterns were expressed via grey-scale. The grey-scale co-occurrence matrixes with four angles of adjacent pixels were used. The texture metrics included the mean and standard deviation of energy, entropy, correlation, inverse different moment, inertia, cluster shade, cluster prominence, and Haralick correlation. Then, the quantitative features were combined in a multinomial logistic regression classifier to generate a prediction model of rotator cuff lesions. Multinomial logistic regression classifier is widely used in the classification of more than two categories such as the three lesion types used in this study. In the classifier, backward elimination was used to select a feature subset which is the most relevant. They were selected from the trained classifier with the lowest error rate. Leave-one-out cross-validation was used to evaluate the performance of the classifier. Each case was left out of the total cases and used to test the trained result by the remaining cases. According to the physician’s assessment, the performance of the proposed CAD system was shown by the accuracy. As a result, the proposed system achieved an accuracy of 86%. A CAD system based on the statistical texture features to interpret echogenicity values in shoulder musculoskeletal ultrasound was established to generate a prediction model for rotator cuff lesions. Clinically, it is difficult to distinguish some kinds of rotator cuff lesions, especially partial-thickness tear of rotator cuff. The shoulder orthopaedic surgeon and musculoskeletal radiologist reported greater diagnostic test accuracy than general radiologist or ultrasonographers based on the available literature. Consequently, the proposed CAD system which was developed according to the experiment of the shoulder orthopaedic surgeon can provide reliable suggestions to general radiologists or ultrasonographers. More quantitative features related to the specific patterns of different lesion types would be investigated in the further study to improve the prediction.

Keywords: shoulder ultrasound, rotator cuff lesions, texture, computer-aided diagnosis

Procedia PDF Downloads 284

Authors: Zunaira Zaman, David Bohannon

Abstract:

This study explores the application of machine learning (ML) for detecting security vulnerabilities in source code. The research aims to assist organizations with large application portfolios and limited security testing capabilities in prioritizing security activities. ML-based approaches offer benefits such as increased confidence scores, false positives and negatives tuning, and automated feedback. The initial approach using natural language processing techniques to extract features achieved 86% accuracy during the training phase but suffered from overfitting and performed poorly on unseen datasets during testing. To address these issues, the study proposes using the abstract syntax tree (AST) for Java and C++ codebases to capture code semantics and structure and generate path-context representations for each function. The Code2Vec model architecture is used to learn distributed representations of source code snippets for training a machine-learning classifier for vulnerability prediction. The study evaluates the performance of the proposed methodology using two datasets and compares the results with existing approaches. The Devign dataset yielded 60% accuracy in predicting vulnerable code snippets and helped resist overfitting, while the Juliet Test Suite predicted specific vulnerabilities such as OS-Command Injection, Cryptographic, and Cross-Site Scripting vulnerabilities. The Code2Vec model achieved 75% accuracy and a 98% recall rate in predicting OS-Command Injection vulnerabilities. The study concludes that even partial AST representations of source code can be useful for vulnerability prediction. The approach has the potential for automated intelligent analysis of source code, including vulnerability prediction on unseen source code. State-of-the-art models using natural language processing techniques and CNN models with ensemble modelling techniques did not generalize well on unseen data and faced overfitting issues. However, predicting vulnerabilities in source code using machine learning poses challenges such as high dimensionality and complexity of source code, imbalanced datasets, and identifying specific types of vulnerabilities. Future work will address these challenges and expand the scope of the research.

Keywords: code embeddings, neural networks, natural language processing, OS command injection, software security, code properties

Procedia PDF Downloads 107

Authors: D. Pugazhenthi, B. Sree Vidya

Abstract:

Cloud computing is one of the emerging technologies that enables end users to use the services of cloud on ‘pay per usage’ strategy. This technology grows in a fast pace and so is its security threat. One among the various services provided by cloud is storage. In this service, security plays a vital factor for both authenticating legitimate users and protection of information. This paper brings in efficient ways of authenticating users as well as securing information on the cloud. Initial phase proposed in this paper deals with an authentication technique using multi-factor and multi-dimensional authentication system with multi-level security. Unique identification and slow intrusive formulates an advanced reliability on user-behaviour based biometrics than conventional means of password authentication. By biometric systems, the accounts are accessed only by a legitimate user and not by a nonentity. The biometric templates employed here do not include single trait but multiple, viz., iris and finger prints. The coordinating stage of the authentication system functions on Ensemble Support Vector Machine (SVM) and optimization by assembling weights of base SVMs for SVM ensemble after individual SVM of ensemble is trained by the Artificial Fish Swarm Algorithm (AFSA). Thus it helps in generating a user-specific secure cryptographic key of the multimodal biometric template by fusion process. Data security problem is averted and enhanced security architecture is proposed using encryption and decryption system with double key cryptography based on Fuzzy Neural Network (FNN) for data storing and retrieval in cloud computing . The proposing scheme aims to protect the records from hackers by arresting the breaking of cipher text to original text. This improves the authentication performance that the proposed double cryptographic key scheme is capable of providing better user authentication and better security which distinguish between the genuine and fake users. Thus, there are three important modules in this proposed work such as 1) Feature extraction, 2) Multimodal biometric template generation and 3) Cryptographic key generation. The extraction of the feature and texture properties from the respective fingerprint and iris images has been done initially. Finally, with the help of fuzzy neural network and symmetric cryptography algorithm, the technique of double key encryption technique has been developed. As the proposed approach is based on neural networks, it has the advantage of not being decrypted by the hacker even though the data were hacked already. The results prove that authentication process is optimal and stored information is secured.

Keywords: artificial fish swarm algorithm (AFSA), biometric authentication, decryption, encryption, fingerprint, fusion, fuzzy neural network (FNN), iris, multi-modal, support vector machine classification

Procedia PDF Downloads 259

Authors: Arth J. Shah, Madhu R. Kamble

Abstract:

Voice biometric liveness detection is customized to certify an authentication process of the voice data presented is genuine and not a recording or synthetic voice. With the rise of deepfakes and other equivalently sophisticated spoofing generation techniques, it’s becoming challenging to ensure that the person on the other end is a live speaker or not. Voice Liveness Detection (VLD) system is a group of security measures which detect and prevent voice spoofing attacks. Motivated by the recent development of the Kolmogorov-Arnold Network (KAN) based on the Kolmogorov-Arnold theorem, we proposed KAN for the VLD task. To date, multilayer perceptron (MLP) based classifiers have been used for the classification tasks. We aim to capture not only the compositional structure of the model but also to optimize the values of univariate functions. This study explains the mathematical as well as experimental analysis of KAN for VLD tasks, thereby opening a new perspective for scientists to work on speech and signal processing-based tasks. This study emerges as a combination of traditional signal processing tasks and new deep learning models, which further proved to be a better combination for VLD tasks. The experiments are performed on the POCO and ASVSpoof 2017 V2 database. We used Constant Q-transform, Mel, and short-time Fourier transform (STFT) based front-end features and used CNN, BiLSTM, and KAN as back-end classifiers. The best accuracy is 91.26 % on the POCO database using STFT features with the KAN classifier. In the ASVSpoof 2017 V2 database, the lowest EER we obtained was 26.42 %, using CQT features and KAN as a classifier.

Keywords: Kolmogorov Arnold networks, multilayer perceptron, pop noise, voice liveness detection

Procedia PDF Downloads 41

Authors: Lae-Jeong Park

Abstract:

The paper presents a method that utilizes figure-ground color segmentation to extract effective global feature in terms of false positive reduction in the head-shoulder detection. Conventional detectors that rely on local features such as HOG due to real-time operation suffer from false positives. Color cue in an input image provides salient information on a global characteristic which is necessary to alleviate the false positives of the local feature based detectors. An effective approach that uses figure-ground color segmentation has been presented in an effort to reduce the false positives in object detection. In this paper, an extended version of the approach is presented that adopts separate multipart foregrounds instead of a single prior foreground and performs the figure-ground color segmentation with each of the foregrounds. The multipart foregrounds include the parts of the head-shoulder shape and additional auxiliary foregrounds being optimized by a search algorithm. A classifier is constructed with the feature that consists of a set of the multiple resulting segmentations. Experimental results show that the presented method can discriminate more false positive than the single prior shape-based classifier as well as detectors with the local features. The improvement is possible because the presented approach can reduce the false positives that have the same colors in the head and shoulder foregrounds.

Keywords: pedestrian detection, color segmentation, false positive, feature extraction

Procedia PDF Downloads 281

Authors: Gayathri Nagarajan, L. D. Dhinesh Babu

Abstract:

Health care is one of the prominent industries that generate voluminous data thereby finding the need of machine learning techniques with big data solutions for efficient processing and prediction. Missing data, incomplete data, real time streaming data, sensitive data, privacy, heterogeneity are few of the common challenges to be addressed for efficient processing and mining of health care data. In comparison with other applications, accuracy and fast processing are of higher importance for health care applications as they are related to the human life directly. Though there are many machine learning techniques and big data solutions used for efficient processing and prediction in health care data, different techniques and different frameworks are proved to be effective for different applications largely depending on the characteristics of the datasets. In this paper, we present a framework that uses ensemble machine learning technique gradient boosted trees for data classification in health care big data. The framework is built on Spark platform which is fast in comparison with other traditional frameworks. Unlike other works that focus on a single technique, our work presents a comparison of six different machine learning techniques along with gradient boosted trees on datasets of different characteristics. Five benchmark health care datasets are considered for experimentation, and the results of different machine learning techniques are discussed in comparison with gradient boosted trees. The metric chosen for comparison is misclassification error rate and the run time of the algorithms. The goal of this paper is to i) Compare the performance of gradient boosted trees with other machine learning techniques in Spark platform specifically for health care big data and ii) Discuss the results from the experiments conducted on datasets of different characteristics thereby drawing inference and conclusion. The experimental results show that the accuracy is largely dependent on the characteristics of the datasets for other machine learning techniques whereas gradient boosting trees yields reasonably stable results in terms of accuracy without largely depending on the dataset characteristics.

Keywords: big data analytics, ensemble machine learning, gradient boosted trees, Spark platform

Procedia PDF Downloads 241

Authors: Armin Bodaghkhani, Bruce Colbourne, Yuri S. Muzychka

Abstract:

The process of spray-cloud formation and flow kinematics produced from breaking wave impact on vertical and slant lab-scale bow-shaped models were experimentally investigated. Bubble Image Velocimetry (BIV) and Image Processing (IP) techniques were applied to study the various types of wave-model impacts. Different wave characteristics were generated in a tow tank to investigate the effects of wave characteristics, such as wave phase velocity, wave steepness on droplet velocities, and behavior of the process of spray cloud formation. The phase ensemble-averaged vertical velocity and turbulent intensity were computed. A high-speed camera and diffused LED backlights were utilized to capture images for further post processing. Various pressure sensors and capacitive wave probes were used to measure the wave impact pressure and the free surface profile at different locations of the model and wave-tank, respectively. Droplet sizes and velocities were measured using BIV and IP techniques to trace bubbles and droplets in order to measure their velocities and sizes by correlating the texture in these images. The impact pressure and droplet size distributions were compared to several previously experimental models, and satisfactory agreements were achieved. The distribution of droplets in front of both models are demonstrated. Due to the highly transient process of spray formation, the drag coefficient for several stages of this transient displacement for various droplet size ranges and different Reynolds number were calculated based on the ensemble average method. From the experimental results, the slant model produces less spray in comparison with the vertical model, and the droplet velocities generated from the wave impact with the slant model have a lower velocity as compared with the vertical model.

Keywords: spray charachteristics, droplet size and velocity, wave-body interactions, bubble image velocimetry, image processing

Procedia PDF Downloads 300

Authors: Kumar Shashvat, Amol P. Bhondekar

Abstract:

In the midst of the five senses, odor is the most reminiscent and least understood. Odor testing has been mysterious and odor data fabled to most practitioners. The delinquent of recognition and classification of odor is important to achieve. The facility to smell and predict whether the artifact is of further use or it has become undesirable for consumption; the imitation of this problem hooked on a model is of consideration. The general industrial standard for this classification is color based anyhow; odor can be improved classifier than color based classification and if incorporated in machine will be awfully constructive. For cataloging of odor for peas, trees and cashews various discriminative approaches have been used Discriminative approaches offer good prognostic performance and have been widely used in many applications but are incapable to make effectual use of the unlabeled information. In such scenarios, generative approaches have better applicability, as they are able to knob glitches, such as in set-ups where variability in the series of possible input vectors is enormous. Generative models are integrated in machine learning for either modeling data directly or as a transitional step to form an indeterminate probability density function. The algorithms or models Linear Discriminant Analysis and Naive Bayes Classifier have been used for classification of the odor of cashews. Linear Discriminant Analysis is a method used in data classification, pattern recognition, and machine learning to discover a linear combination of features that typifies or divides two or more classes of objects or procedures. The Naive Bayes algorithm is a classification approach base on Bayes rule and a set of qualified independence theory. Naive Bayes classifiers are highly scalable, requiring a number of restraints linear in the number of variables (features/predictors) in a learning predicament. The main recompenses of using the generative models are generally a Generative Models make stronger assumptions about the data, specifically, about the distribution of predictors given the response variables. The Electronic instrument which is used for artificial odor sensing and classification is an electronic nose. This device is designed to imitate the anthropological sense of odor by providing an analysis of individual chemicals or chemical mixtures. The experimental results have been evaluated in the form of the performance measures i.e. are accuracy, precision and recall. The investigational results have proven that the overall performance of the Linear Discriminant Analysis was better in assessment to the Naive Bayes Classifier on cashew dataset.

Keywords: odor classification, generative models, naive bayes, linear discriminant analysis

Procedia PDF Downloads 388

Authors: Rohit Gupta, Ravinder Agarwal

Abstract:

Surface electromyographic (sEMG) signal has the potential to identify the human activities and intention. This potential is further exploited to control the artificial limbs using the sEMG signal from residual limbs of amputees. The paper deals with the development of multichannel cost efficient sEMG signal interface for research application, along with evaluation of proposed class dependent statistical approach of the feature selection method. The sEMG signal acquisition interface was developed using ADS1298 of Texas Instruments, which is a front-end interface integrated circuit for ECG application. Further, the sEMG signal is recorded from two lower limb muscles for three locomotions namely: Plane Walk (PW), Stair Ascending (SA), Stair Descending (SD). A class dependent statistical approach is proposed for feature selection and also its performance is compared with 12 preexisting feature vectors. To make the study more extensive, performance of five different types of classifiers are compared. The outcome of the current piece of work proves the suitability of the proposed feature selection algorithm for locomotion recognition, as compared to other existing feature vectors. The SVM Classifier is found as the outperformed classifier among compared classifiers with an average recognition accuracy of 97.40%. Feature vector selection emerges as the most dominant factor affecting the classification performance as it holds 51.51% of the total variance in classification accuracy. The results demonstrate the potentials of the developed sEMG signal acquisition interface along with the proposed feature selection algorithm.

Keywords: classifiers, feature selection, locomotion, sEMG

Procedia PDF Downloads 293

Authors: Azita Ramezani, Ghazal Mashhadiagha, Bahareh Sanabakhsh

Abstract:

This study researches the combination of Random. Forest classifiers with large language models (LLMs) and natural language processing (NLP) to improve diagnostic accuracy in chest X-ray analysis using the Indiana University dataset. Utilizing advanced NLP techniques, the research preprocesses textual data from radiological reports to extract key features, which are then merged with image-derived data. This improved dataset is analyzed with Random Forest classifiers to predict specific clinical results, focusing on the identification of health issues and the estimation of case urgency. The findings reveal that the combination of NLP, LLMs, and machine learning not only increases diagnostic precision but also reliability, especially in quickly identifying critical conditions. Achieving an accuracy of 99.35%, the model shows significant advancements over conventional diagnostic techniques. The results emphasize the large potential of machine learning in medical imaging, suggesting that these technologies could greatly enhance clinician judgment and patient outcomes by offering quicker and more precise diagnostic approximations.

Keywords: natural language processing (NLP), large language models (LLMs), random forest classifier, chest x-ray analysis, medical imaging, diagnostic accuracy, indiana university dataset, machine learning in healthcare, predictive modeling, clinical decision support systems

Procedia PDF Downloads 46

Authors: Elena Collino, Dario A. Ronzio, Goffredo Decimi, Maurizio Riva

Abstract:

The rapid increase of renewable energy in Italy is led by wind and solar installations. The 2017 Italian energy strategy foresees a further development of these sustainable technologies, especially solar. This fact has resulted in new opportunities, challenges, and different problems to deal with. The growth of renewables allows to meet the European requirements regarding energy and environmental policy, but these types of sources are difficult to manage because they are intermittent and non-programmable. Operationally, these characteristics can lead to instability on the voltage profile and increasing uncertainty on energy reserve scheduling. The increasing renewable production must be considered with more and more attention especially by the Transmission System Operator (TSO). The TSO, in fact, every day provides orders on energy dispatch, once the market outcome has been determined, on extended areas, defined mainly on the basis of power transmission limitations. In Italy, six market zone are defined: Northern-Italy, Central-Northern Italy, Central-Southern Italy, Southern Italy, Sardinia, and Sicily. An accurate hourly renewable power forecasting for the day-ahead on these extended areas brings an improvement both in terms of dispatching and reserve management. In this study, an operational forecasting tool of the hourly solar output for the six Italian market zones is presented, and the performance is analysed. The implementation is carried out by means of a numerical weather prediction model, coupled with a statistical post-processing in order to derive the power forecast on the basis of the meteorological projection. The weather forecast is obtained from the limited area model RAMS on the Italian territory, initialized with IFS-ECMWF boundary conditions. The post-processing calculates the solar power production with the Analog Ensemble technique (AN). This statistical approach forecasts the production using a probability distribution of the measured production registered in the past when the weather scenario looked very similar to the forecasted one. The similarity is evaluated for the components of the solar radiation: global (GHI), diffuse (DIF) and direct normal (DNI) irradiation, together with the corresponding azimuth and zenith solar angles. These are, in fact, the main factors that affect the solar production. Considering that the AN performance is strictly related to the length and quality of the historical data a training period of more than one year has been used. The training set is made by historical Numerical Weather Prediction (NWP) forecasts at 12 UTC for the GHI, DIF and DNI variables over the Italian territory together with corresponding hourly measured production for each of the six zones. The AN technique makes it possible to estimate the aggregate solar production in the area, without information about the technologic characteristics of the all solar parks present in each area. Besides, this information is often only partially available. Every day, the hourly solar power forecast for the six Italian market zones is made publicly available through a website.

Keywords: analog ensemble, electricity market, PV forecast, solar energy

Procedia PDF Downloads 158

Authors: Asma Bentaifa

Abstract:

This paper investigates the effect of ownership concentration and especially the presence of controlling shareholders on the firm’s payout decisions. Using a sample of 870 French companies during 2007 to 2012, we find that the share of dividends in total payout is negatively correlated with the size of cash flow held by controlling shareholder, and positively related to the divergence between voting rights and cash flow rights of largest shareholders. We also document that controlled firms tend to prefer dividends over repurchases to mitigate conflicts between controlling shareholders and minority shareholders related to the presence of control enhancing devices.

Keywords: ownership, payout policy, dividend, minority expropriation

Procedia PDF Downloads 222

Authors: Madushani Rodrigo, Banuka Athuraliya

Abstract:

In today's world of medical diagnosis and prediction, machine learning stands out as a strong tool, transforming old ways of caring for health. This study analyzes the use of machine learning in the specialized domain of sports medicine, with a focus on the timely and accurate detection of bone fractures in cricket athletes. Failure to identify bone fractures in real time can result in malunion or non-union conditions. To ensure proper treatment and enhance the bone healing process, accurately identifying fracture locations and types is necessary. When interpreting X-ray images, it relies on the expertise and experience of medical professionals in the identification process. Sometimes, radiographic images are of low quality, leading to potential issues. Therefore, it is necessary to have a proper approach to accurately localize and classify fractures in real time. The research has revealed that the optimal approach needs to address the stated problem and employ appropriate radiographic image processing techniques and object detection algorithms. These algorithms should effectively localize and accurately classify all types of fractures with high precision and in a timely manner. In order to overcome the challenges of misidentifying fractures, a distinct model for fracture localization and classification has been implemented. The research also incorporates radiographic image enhancement and preprocessing techniques to overcome the limitations posed by low-quality images. A classification ensemble model has been implemented using ResNet18 and VGG16. In parallel, a fracture segmentation model has been implemented using the enhanced U-Net architecture. Combining the results of these two implemented models, the FracXpert system can accurately localize exact fracture locations along with fracture types from the available 12 different types of fracture patterns, which include avulsion, comminuted, compressed, dislocation, greenstick, hairline, impacted, intraarticular, longitudinal, oblique, pathological, and spiral. This system will generate a confidence score level indicating the degree of confidence in the predicted result. Using ResNet18 and VGG16 architectures, the implemented fracture segmentation model, based on the U-Net architecture, achieved a high accuracy level of 99.94%, demonstrating its precision in identifying fracture locations. Simultaneously, the classification ensemble model achieved an accuracy of 81.0%, showcasing its ability to categorize various fracture patterns, which is instrumental in the fracture treatment process. In conclusion, FracXpert has become a promising ML application in sports medicine, demonstrating its potential to revolutionize fracture detection processes. By leveraging the power of ML algorithms, this study contributes to the advancement of diagnostic capabilities in cricket athlete healthcare, ensuring timely and accurate identification of bone fractures for the best treatment outcomes.

Keywords: multiclass classification, object detection, ResNet18, U-Net, VGG16

Procedia PDF Downloads 122

Authors: M. Kharrat, G. Moreau, Z. Aboura

Abstract:

The latest advances in the weaving industry, combined with increasingly sophisticated means of materials processing, have made it possible to produce complex 3D composite structures. Mainly used in aeronautics, composite materials with 3D architecture offer better mechanical properties than 2D reinforced composites. Nevertheless, these materials require a good understanding of their behavior. Because of the complexity of such materials, the damage mechanisms are multiple, and the scenario of their appearance and evolution depends on the nature of the exerted solicitations. The AE technique is a well-established tool for discriminating between the damage mechanisms. Suitable sensors are used during the mechanical test to monitor the structural health of the material. Relevant AE-features are then extracted from the recorded signals, followed by a data analysis using pattern recognition techniques. In order to better understand the damage scenarios of interlock composite materials, a multi-instrumentation was set-up in this work for tracking damage initiation and development, especially in the vicinity of the first significant damage, called macro-damage. The deployed instrumentation includes video-microscopy, Digital Image Correlation, Acoustic Emission (AE) and micro-tomography. In this study, a multi-variable AE data analysis approach was developed for the discrimination between the different signal classes representing the different emission sources during testing. An unsupervised classification technique was adopted to perform AE data clustering without a priori knowledge. The multi-instrumentation and the clustered data served to label the different signal families and to build a learning database. This latter is useful to construct a supervised classifier that can be used for automatic recognition of the AE signals. Several materials with different ingredients were tested under various solicitations in order to feed and enrich the learning database. The methodology presented in this work was useful to refine the damage threshold for the new generation materials. The damage mechanisms around this threshold were highlighted. The obtained signal classes were assigned to the different mechanisms. The isolation of a 'noise' class makes it possible to discriminate between the signals emitted by damages without resorting to spatial filtering or increasing the AE detection threshold. The approach was validated on different material configurations. For the same material and the same type of solicitation, the identified classes are reproducible and little disturbed. The supervised classifier constructed based on the learning database was able to predict the labels of the classified signals.

Keywords: acoustic emission, classifier, damage mechanisms, first damage threshold, interlock composite materials, pattern recognition

Procedia PDF Downloads 155

Authors: Areej Salaymeh, Loren Schwiebert, Stephen Remias, Jonathan Waddell

Abstract:

Extracting information from new data sources has emerged as a crucial task in many traffic planning processes, such as identifying traffic patterns, route planning, traffic forecasting, and locating infrastructure improvements. Given the advanced technologies used to collect Global Positioning System (GPS) data from dedicated GPS devices, GPS equipped phones, and navigation tools, intelligent data analysis methodologies are necessary to mine this raw data. In this research, an automatic detection framework is proposed to help identify and classify the locations of stopped GPS waypoints into two main categories: signalized intersections or highway congestion. The Delaunay triangulation is used to perform this assessment in the clustering phase. While most of the existing clustering algorithms need assumptions about the data distribution, the effectiveness of the Delaunay triangulation relies on triangulating geographical data points without such assumptions. Our proposed method starts by cleaning noise from the data and normalizing it. Next, the framework will identify stoppage points by calculating the traveled distance. The last step is to use clustering to form groups of waypoints for signalized traffic and highway congestion. Next, a binary classifier was applied to find distinguish highway congestion from signalized stop points. The binary classifier uses the length of the cluster to find congestion. The proposed framework shows high accuracy for identifying the stop positions and congestion points in around 99.2% of trials. We show that it is possible, using limited GPS data, to distinguish with high accuracy.

Keywords: Delaunay triangulation, clustering, intelligent transportation systems, GPS data

Procedia PDF Downloads 275

Authors: A: Annis Fathima, V. Vaidehi, S. Ajitha

Abstract:

Face recognition system finds many applications in surveillance and human computer interaction systems. As the applications using face recognition systems are of much importance and demand more accuracy, more robustness in the face recognition system is expected with less computation time. In this paper, a hybrid approach for face recognition combining Gabor Wavelet and Linear Discriminant Analysis (HGWLDA) is proposed. The normalized input grayscale image is approximated and reduced in dimension to lower the processing overhead for Gabor filters. This image is convolved with bank of Gabor filters with varying scales and orientations. LDA, a subspace analysis techniques are used to reduce the intra-class space and maximize the inter-class space. The techniques used are 2-dimensional Linear Discriminant Analysis (2D-LDA), 2-dimensional bidirectional LDA ((2D)2LDA), Weighted 2-dimensional bidirectional Linear Discriminant Analysis (Wt (2D)2 LDA). LDA reduces the feature dimension by extracting the features with greater variance. k-Nearest Neighbour (k-NN) classifier is used to classify and recognize the test image by comparing its feature with each of the training set features. The HGWLDA approach is robust against illumination conditions as the Gabor features are illumination invariant. This approach also aims at a better recognition rate using less number of features for varying expressions. The performance of the proposed HGWLDA approaches is evaluated using AT&T database, MIT-India face database and faces94 database. It is found that the proposed HGWLDA approach provides better results than the existing Gabor approach.

Keywords: face recognition, Gabor wavelet, LDA, k-NN classifier

Procedia PDF Downloads 467

Authors: Yanjie Zhu, André Jesus, Irwanda Laory

Abstract:

Structural Health Monitoring (SHM), involving data acquisition, data interpretation and decision-making system aim to continuously monitor the structural performance of civil infrastructures under various in-service circumstances. The main value and purpose of SHM is identifying damages through data interpretation system. Research on SHM has been expanded in the last decades and a large volume of data is recorded every day owing to the dramatic development in sensor techniques and certain progress in signal processing techniques. However, efficient and reliable data interpretation for damage detection under environmental variations is still a big challenge. Structural damages might be masked because variations in measured data can be the result of environmental variations. This research reports a novel method based on single-channel Blind Signal Separation (BSS), which extracts environmental effects from measured data directly without any prior knowledge of the structure loading and environmental conditions. Despite the successful application in audio processing and bio-medical research fields, BSS has never been used to detect damage under varying environmental conditions. This proposed method optimizes and combines Ensemble Empirical Mode Decomposition (EEMD), Principal Component Analysis (PCA) and Independent Component Analysis (ICA) together to separate structural responses due to different loading conditions respectively from a single channel input signal. The ICA is applying on dimension-reduced output of EEMD. Numerical simulation of a truss bridge, inspired from New Joban Line Arakawa Railway Bridge, is used to validate this method. All results demonstrate that the single-channel BSS-based method can recover temperature effects from mixed structural response recorded by a single sensor with a convincing accuracy. This will be the foundation of further research on direct damage detection under varying environment.

Keywords: damage detection, ensemble empirical mode decomposition (EEMD), environmental variations, independent component analysis (ICA), principal component analysis (PCA), structural health monitoring (SHM)

Procedia PDF Downloads 304

Authors: Sifiso Vilakazi, Lerato Mzenzi

Abstract:

The incorporation of technology has brought about a considerable evolution in election management, providing answers to persistent issues with fraud, inefficiency, and logistical complexity. The paper offers a comparative review of the effects of technology advancements on election logistics and fraud prevention in Uganda and Nigeria. Both nations have embraced technology such as digital fraud prevention systems, biometric voter registration, and electronic results transmission while having different political environments and electoral problems. Nevertheless, these innovations' varied results raise important concerns about how technology can enhance vote integrity. For improved transparency and lower voter fraud, the Independent National Electoral Commission (INEC) of Nigeria has deployed electronic voting machines, biometric voter identification, and the INEC Result Viewing (IReV) portal. Despite these developments, technological setbacks and logistical difficulties, particularly during the 2023 elections, uncovered weaknesses that stoked distrust and electoral conflicts by exposing flaws, including device breakdowns, insufficient cybersecurity protections, and transmission delays in results. Comparably, Uganda has used electronic result transmission technologies and biometric voter verification since 2016 to increase election efficiency and combat ballot stuffing and impersonation. Nevertheless, problems, including remote logistical challenges and internet outages during the 2021 elections, have reduced the efficacy of these tools. This paper maintains that while technology might reduce election-related stress and potentially reduce fraud, its efficacy depends on several variables, such as political will, public confidence, and infrastructure. Furthermore, it draws attention to the necessity of more robust legislative frameworks, ongoing investments in cybersecurity, and implementation plans that are customized to the particular difficulties presented by each nation's voting system locally. The results imply that although technology can help Ugandan and Nigerian election management, it cannot guarantee electoral integrity and must be used in conjunction with more extensive institutional changes. Through providing insights into how African nations might use technological advancements to improve democratic governance while addressing context-specific problems, the research adds to the expanding body of literature on the use of technology in election management.

Keywords: elections, Nigeria, Uganda, Africa, management, innovation

Procedia PDF Downloads 6

Authors: Bliss Singhal

Abstract:

Machine learning (ML) is a branch of Artificial Intelligence (AI) where computers analyze data and find patterns in the data. The study focuses on the detection of metastatic cancer using ML. Metastatic cancer is the stage where cancer has spread to other parts of the body and is the cause of approximately 90% of cancer-related deaths. Normally, pathologists spend hours each day to manually classifying whether tumors are benign or malignant. This tedious task contributes to mislabeling metastasis being over 60% of the time and emphasizes the importance of being aware of human error and other inefficiencies. ML is a good candidate to improve the correct identification of metastatic cancer, saving thousands of lives and can also improve the speed and efficiency of the process, thereby taking fewer resources and time. So far, the deep learning methodology of AI has been used in research to detect cancer. This study is a novel approach to determining the potential of using preprocessing algorithms combined with classification algorithms in detecting metastatic cancer. The study used two preprocessing algorithms: principal component analysis (PCA) and the genetic algorithm, to reduce the dimensionality of the dataset and then used three classification algorithms: logistic regression, decision tree classifier, and k-nearest neighbors to detect metastatic cancer in the pathology scans. The highest accuracy of 71.14% was produced by the ML pipeline comprising of PCA, the genetic algorithm, and the k-nearest neighbor algorithm, suggesting that preprocessing and classification algorithms have great potential for detecting metastatic cancer.

Keywords: breast cancer, principal component analysis, genetic algorithm, k-nearest neighbors, decision tree classifier, logistic regression

Procedia PDF Downloads 82

Authors: Asma Omri, Noureddine Benothman, Sofiane Sayahi, Fethi Tlili, Hichem Besbes

Abstract:

Cycling, a popular mode of transportation, can also be perilous due to cyclists' vulnerability to collisions with vehicles and obstacles. This paper presents an innovative cyclist safety system based on radar technology designed to offer real-time collision risk warnings to cyclists. The system incorporates a low-power radar sensor affixed to the bicycle and connected to a microcontroller. It leverages radar point cloud detections, a clustering algorithm, and a supervised classifier. These algorithms are optimized for efficiency to run on the TI’s AWR 1843 BOOST radar, utilizing a coarse classification approach distinguishing between cars, trucks, two-wheeled vehicles, and other objects. To enhance the performance of clustering techniques, we propose a 2-Level clustering approach. This approach builds on the state-of-the-art Density-based spatial clustering of applications with noise (DBSCAN). The objective is to first cluster objects based on their velocity, then refine the analysis by clustering based on position. The initial level identifies groups of objects with similar velocities and movement patterns. The subsequent level refines the analysis by considering the spatial distribution of these objects. The clusters obtained from the first level serve as input for the second level of clustering. Our proposed technique surpasses the classical DBSCAN algorithm in terms of geometrical metrics, including homogeneity, completeness, and V-score. Relevant cluster features are extracted and utilized to classify objects using an SVM classifier. Potential obstacles are identified based on their velocity and proximity to the cyclist. To optimize the system, we used the View of Delft dataset for hyperparameter selection and SVM classifier training. The system's performance was assessed using our collected dataset of radar point clouds synchronized with a camera on an Nvidia Jetson Nano board. The radar-based cyclist safety system is a practical solution that can be easily installed on any bicycle and connected to smartphones or other devices, offering real-time feedback and navigation assistance to cyclists. We conducted experiments to validate the system's feasibility, achieving an impressive 85% accuracy in the classification task. This system has the potential to significantly reduce the number of accidents involving cyclists and enhance their safety on the road.

Keywords: 2-level clustering, coarse classification, cyclist safety, warning system based on radar technology

Procedia PDF Downloads 80

Authors: Roza Dzierzak, Waldemar Wojcik, Piotr Kacejko

Abstract:

Analysis of the texture of medical images consists of determining the parameters and characteristics of the examined tissue. The main goal is to assign the analyzed area to one of two basic groups: as a healthy tissue or a tissue with pathological changes. The CT images of the thoracic lumbar spine from 15 healthy patients and 15 with confirmed osteoporosis were used for the analysis. As a result, 120 samples with dimensions of 50x50 pixels were obtained. The set of features has been obtained based on the histogram, gradient, run-length matrix, co-occurrence matrix, autoregressive model, and Haar wavelet. As a result of the image analysis, 290 descriptors of textural features were obtained. The dimension of the space of features was reduced by the use of three selection methods: Fisher coefficient (FC), mutual information (MI), minimization of the classification error probability and average correlation coefficients between the chosen features minimization of classification error probability (POE) and average correlation coefficients (ACC). Each of them returned ten features occupying the initial place in the ranking devised according to its own coefficient. As a result of the Fisher coefficient and mutual information selections, the same features arranged in a different order were obtained. In both rankings, the 50% percentile (Perc.50%) was found in the first place. The next selected features come from the co-occurrence matrix. The sets of features selected in the selection process were evaluated using six classification tree methods. These were: decision stump (DS), Hoeffding tree (HT), logistic model trees (LMT), random forest (RF), random tree (RT) and reduced error pruning tree (REPT). In order to assess the accuracy of classifiers, the following parameters were used: overall classification accuracy (ACC), true positive rate (TPR, classification sensitivity), true negative rate (TNR, classification specificity), positive predictive value (PPV) and negative predictive value (NPV). Taking into account the classification results, it should be stated that the best results were obtained for the Hoeffding tree and logistic model trees classifiers, using the set of features selected by the POE + ACC method. In the case of the Hoeffding tree classifier, the highest values of three parameters were obtained: ACC = 90%, TPR = 93.3% and PPV = 93.3%. Additionally, the values of the other two parameters, i.e., TNR = 86.7% and NPV = 86.6% were close to the maximum values obtained for the LMT classifier. In the case of logistic model trees classifier, the same ACC value was obtained ACC=90% and the highest values for TNR=88.3% and NPV= 88.3%. The values of the other two parameters remained at a level close to the highest TPR = 91.7% and PPV = 91.6%. The results obtained in the experiment show that the use of classification trees is an effective method of classification of texture features. This allows identifying the conditions of the spongy tissue for healthy cases and those with the porosis.

Keywords: classification, feature selection, texture analysis, tree algorithms

Procedia PDF Downloads 178

Authors: Qiaosi Li, Frankie Kwan Kit Wong, Tung Fung

Abstract:

Lightweight unmanned aerial vehicle (UAV) loading with novel sensors offers a low cost approach for data acquisition in complex environment. This study established a framework for applying UAV system in complex environment quick mapping and assessed the performance of UAV-based hyperspectral image and digital surface model (DSM) derived from photogrammetric point clouds for 13 species classification in wetland area Mai Po Inner Deep Bay Ramsar Site, Hong Kong. The study area was part of shallow bay with flat terrain and the major species including reedbed and four mangroves: Kandelia obovata, Aegiceras corniculatum, Acrostichum auerum and Acanthus ilicifolius. Other species involved in various graminaceous plants, tarbor, shrub and invasive species Mikania micrantha. In particular, invasive species climbed up to the mangrove canopy caused damage and morphology change which might increase species distinguishing difficulty. Hyperspectral images were acquired by Headwall Nano sensor with spectral range from 400nm to 1000nm and 0.06m spatial resolution image. A sequence of multi-view RGB images was captured with 0.02m spatial resolution and 75% overlap. Hyperspectral image was corrected for radiative and geometric distortion while high resolution RGB images were matched to generate maximum dense point clouds. Furtherly, a 5 cm grid digital surface model (DSM) was derived from dense point clouds. Multiple feature reduction methods were compared to identify the efficient method and to explore the significant spectral bands in distinguishing different species. Examined methods including stepwise discriminant analysis (DA), support vector machine (SVM) and minimum noise fraction (MNF) transformation. Subsequently, spectral subsets composed of the first 20 most importance bands extracted by SVM, DA and MNF, and multi-source subsets adding extra DSM to 20 spectrum bands were served as input in maximum likelihood classifier (MLC) and SVM classifier to compare the classification result. Classification results showed that feature reduction methods from best to worst are MNF transformation, DA and SVM. MNF transformation accuracy was even higher than all bands input result. Selected bands frequently laid along the green peak, red edge and near infrared. Additionally, DA found that chlorophyll absorption red band and yellow band were also important for species classification. In terms of 3D data, DSM enhanced the discriminant capacity among low plants, arbor and mangrove. Meanwhile, DSM largely reduced misclassification due to the shadow effect and morphological variation of inter-species. In respect to classifier, nonparametric SVM outperformed than MLC for high dimension and multi-source data in this study. SVM classifier tended to produce higher overall accuracy and reduce scattered patches although it costs more time than MLC. The best result was obtained by combining MNF components and DSM in SVM classifier. This study offered a precision species distribution survey solution for inaccessible wetland area with low cost of time and labour. In addition, findings relevant to the positive effect of DSM as well as spectral feature identification indicated that the utility of UAV-borne hyperspectral and photogrammetry deriving 3D data is promising in further research on wetland species such as bio-parameters modelling and biological invasion monitoring.

Keywords: digital surface model (DSM), feature reduction, hyperspectral, photogrammetric point cloud, species mapping, unmanned aerial vehicle (UAV)

Procedia PDF Downloads 257

Authors: Fernanda M. Assef, Maria Teresinha A. Steiner

Abstract:

The aim of this paper is to compare and discuss better classifier algorithm options for credit risk assessment by applying different Machine Learning techniques. Using records from a Brazilian financial institution, this study uses a database of 5,432 companies that are clients of the bank, where 2,600 clients are classified as non-defaulters, 1,551 are classified as defaulters and 1,281 are temporarily defaulters, meaning that the clients are overdue on their payments for up 180 days. For each case, a total of 15 attributes was considered for a one-against-all assessment using four different techniques: Artificial Neural Networks Multilayer Perceptron (ANN-MLP), Artificial Neural Networks Radial Basis Functions (ANN-RBF), Logistic Regression (LR) and finally Support Vector Machines (SVM). For each method, different parameters were analyzed in order to obtain different results when the best of each technique was compared. Initially the data were coded in thermometer code (numerical attributes) or dummy coding (for nominal attributes). The methods were then evaluated for each parameter and the best result of each technique was compared in terms of accuracy, false positives, false negatives, true positives and true negatives. This comparison showed that the best method, in terms of accuracy, was ANN-RBF (79.20% for non-defaulter classification, 97.74% for defaulters and 75.37% for the temporarily defaulter classification). However, the best accuracy does not always represent the best technique. For instance, on the classification of temporarily defaulters, this technique, in terms of false positives, was surpassed by SVM, which had the lowest rate (0.07%) of false positive classifications. All these intrinsic details are discussed considering the results found, and an overview of what was presented is shown in the conclusion of this study.

Keywords: artificial neural networks (ANNs), classifier algorithms, credit risk assessment, logistic regression, machine Learning, support vector machines

Procedia PDF Downloads 103

Authors: Samhita Bharadwaj

Abstract:

This is an empirical study of classifiers in Assamese and Bodo, two genetically unrelated languages of India. The objective of the paper is to address the language contact between Assamese and Bodo as reflected in classifiers. The data has been collected through fieldwork in Bodo recording narratives and folk tales and eliciting specific data from the speakers. The data for Assamese is self-produced as native speaker of the language. Assamese is the easternmost New-Indo-Aryan (henceforth NIA) language mainly spoken in the Brahmaputra valley of Assam and some other north-eastern states of India. It is the lingua franca of Assam and is creolised in the neighbouring state of Nagaland. Bodo, on the other hand, is a Tibeto-Burman (henceforth TB) language of the Bodo-Garo group. It has the highest number of speakers among the TB languages of Assam. However, compared to Assamese, it is still a lesser documented language and due to the prestige of Assamese, all the Bodo speakers are fluent bi-lingual in Assamese, though the opposite isn’t the case. With this context, classifiers, a characteristic phenomenon of TB languages, but not so much of NIA languages, presents an interesting case study on language contact caused by bilingualism. Assamese, as a result of its language contact with the TB languages which are rich in classifiers; has developed the richest classifier system among the IA languages in India. Yet, as a part of rampant borrowing of Assamese words and patterns into Bodo; Bodo is seen to borrow even Assamese classifiers into its system. This paper analyses the borrowed classifiers of Bodo and finds the route of this borrowing phenomenon in the number system of the languages. As the Bodo speakers start replacing the higher numbers from five with Assamese ones, they also choose the Assamese classifiers to attach to these numbers. Thus, the partial loss of number in Bodo as a result of language contact and bilingualism in Assamese is found to be the reason behind the borrowing of classifiers in Bodo. The significance of the study lies in exploring an interesting aspect of language contact in Assam. It is hoped that this will attract further research on bilingualism and classifiers in Assam.

Keywords: Assamese, bi-lingual, Bodo, borrowing, classifier, language contact

Procedia PDF Downloads 223

Authors: Eric Yi-Hsiu Huang, Meng-Chun Kao, Wen-Chuan Kuo

Abstract:

For every laparoscopic surgery, a safe pneumoperitoneumcreation (gaining access to the peritoneal cavity) is the first and essential step. However, closed pneumoperitoneum is usually obtained by blind insertion of a Veress needle into the peritoneal cavity, which may carry potential risks suchas bowel and vascular injury.Until now, there remains no definite measure to visually confirm the position of the needle tip inside the peritoneal cavity. Therefore, this study established an image-guided Veress needle method by combining a fiber probe with optical coherence tomography (OCT). An algorithm was also proposed for determining the exact location of the needle tip through the acquisition of OCT images. Our method not only generates a series of “live” two-dimensional (2D) images during the needle puncture toward the peritoneal cavity but also can eliminate operator variation in image judgment, thus improving peritoneal access safety. This study was approved by the Ethics Committee of Taipei Veterans General Hospital (Taipei VGH IACUC 2020-144). A total of 2400 in vivo OCT images, independent of each other, were acquired from experiments of forty peritoneal punctures on two piglets. Characteristic OCT image patterns could be observed during the puncturing process. The ROC curve demonstrates the discrimination capability of these quantitative image features of the classifier, showing the accuracy of the classifier for determining the inside vs. outside of the peritoneal was 98% (AUC=0.98). In summary, the present study demonstrates the ability of the combination of our proposed automatic identification method and OCT imaging for automatically and objectively identifying the location of the needle tip. OCT images translate the blind closed technique of peritoneal access into a visualized procedure, thus improving peritoneal access safety.

Keywords: pneumoperitoneum, optical coherence tomography, automatic identification, veress needle

Procedia PDF Downloads 134

Authors: Alejandro Molina-Villegas, Silvia Fernández-Sabido, Eduardo Mendoza-Vargas, Fátima Miranda-Pestaña

Abstract:

The automatic detection of indigenous languages ​​in digital texts is essential to promote their inclusion in digital media. Underrepresented languages, such as Maya, are often excluded from language detection tools like Google’s language-detection library, LANGDETECT. This study addresses these limitations by developing a hybrid language detection solution that accurately distinguishes Maya (YUA) from Spanish (ES). Two strategies are employed: the first focuses on creating a profile for the Maya language within the LANGDETECT library, while the second involves training a Naive Bayes classification model with two categories, YUA and ES. The process includes comprehensive data preprocessing steps, such as cleaning, normalization, tokenization, and n-gram counting, applied to text samples collected from various sources, including articles from La Jornada Maya, a major newspaper in Mexico and the only media outlet that includes a Maya section. After the training phase, a portion of the data is used to create the YUA profile within LANGDETECT, which achieves an accuracy rate above 95% in identifying the Maya language during testing. Additionally, the Naive Bayes classifier, trained and tested on the same database, achieves an accuracy close to 98% in distinguishing between Maya and Spanish, with further validation through F1 score, recall, and logarithmic scoring, without signs of overfitting. This strategy, which combines the LANGDETECT profile with a Naive Bayes model, highlights an adaptable framework that can be extended to other underrepresented languages in future research. This fills a gap in Natural Language Processing and supports the preservation and revitalization of these languages.

Keywords: indigenous languages, language detection, Maya language, Naive Bayes classifier, natural language processing, low-resource languages

Procedia PDF Downloads 18

Authors: Bushra Zafar, Usman Qamar

Abstract:

Large data sample size and dimensions render the effectiveness of conventional data mining methodologies. A data mining technique are important tools for collection of knowledgeable information from variety of databases and provides supervised learning in the form of classification to design models to describe vital data classes while structure of the classifier is based on class attribute. Classification efficiency and accuracy are often influenced to great extent by noisy and undesirable features in real application data sets. The inherent natures of data set greatly masks its quality analysis and leave us with quite few practical approaches to use. To our knowledge first time, we present a new approach for investigation of structure and quality of datasets by providing a targeted analysis of localization of noisy and irrelevant features of data sets. Machine learning is based primarily on feature selection as pre-processing step which offers us to select few features from number of features as a subset by reducing the space according to certain evaluation criterion. The primary objective of this study is to trim down the scope of the given data sample by searching a small set of important features which may results into good classification performance. For this purpose, a heuristic for wrapper-based feature selection using genetic algorithm and for discriminative feature selection an external classifier are used. Selection of feature based on its number of occurrence in the chosen chromosomes. Sample dataset has been used to demonstrate proposed idea effectively. A proposed method has improved average accuracy of different datasets is about 95%. Experimental results illustrate that proposed algorithm increases the accuracy of prediction of different diseases.

Keywords: data mining, generic algorithm, KNN algorithms, wrapper based feature selection

Procedia PDF Downloads 316

Authors: Shrabani Ghosh

Abstract:

Automatic offensive language detection in social media has become a stirring task in today's NLP. Manual Offensive language detection is tedious and laborious work where automatic methods based on machine learning are only alternatives. Previous works have done sentiment analysis over social media in different ways such as supervised, semi-supervised, and unsupervised manner. Domain adaptation in a semi-supervised way has also been explored in NLP, where the source domain and the target domain are different. In domain adaptation, the source domain usually has a large amount of labeled data, while only a limited amount of labeled data is available in the target domain. Pretrained transformers like BERT, RoBERTa models are fine-tuned to perform text classification in an unsupervised manner to perform further pre-train masked language modeling (MLM) tasks. In previous work, hate speech detection has been explored in Gab.ai, which is a free speech platform described as a platform of extremist in varying degrees in online social media. In domain adaptation process, Twitter data is used as the source domain, and Gab data is used as the target domain. The performance of domain adaptation also depends on the cross-domain similarity. Different distance measure methods such as L2 distance, cosine distance, Maximum Mean Discrepancy (MMD), Fisher Linear Discriminant (FLD), and CORAL have been used to estimate domain similarity. Certainly, in-domain distances are small, and between-domain distances are expected to be large. The previous work finding shows that pretrain masked language model (MLM) fine-tuned with a mixture of posts of source and target domain gives higher accuracy. However, in-domain performance of the hate classifier on Twitter data accuracy is 71.78%, and out-of-domain performance of the hate classifier on Gab data goes down to 56.53%. Recently self-supervised learning got a lot of attention as it is more applicable when labeled data are scarce. Few works have already been explored to apply self-supervised learning on NLP tasks such as sentiment classification. Self-supervised language representation model ALBERTA focuses on modeling inter-sentence coherence and helps downstream tasks with multi-sentence inputs. Self-supervised attention learning approach shows better performance as it exploits extracted context word in the training process. In this work, a self-supervised attention mechanism has been proposed to detect hate speech on Gab.ai. This framework initially classifies the Gab dataset in an attention-based self-supervised manner. On the next step, a semi-supervised classifier trained on the combination of labeled data from the first step and unlabeled data. The performance of the proposed framework will be compared with the results described earlier and also with optimized outcomes obtained from different optimization techniques.

Keywords: attention learning, language model, offensive language detection, self-supervised learning

Procedia PDF Downloads 106