Search results for: classification accuracy

Authors: Sarah Beaver, Renee Bryce

Abstract:

Audiograms detect hearing impairment, but missing values pose problems. This work explores imputations in an attempt to improve accuracy. This work implements Linear Regression, Lasso, Linear Support Vector Regression, Bayesian Ridge, K Nearest Neighbors (KNN), and Random Forest machine learning techniques to impute audiogram frequencies ranging from 125Hz to 8000Hz. The data contains patients who had or were candidates for cochlear implants. Accuracy is compared across two different Nested Cross-Validation k values. Over 4000 audiograms were used from 800 unique patients. Additionally, training on data combines and compares left and right ear audiograms versus single ear side audiograms. The accuracy achieved using Root Mean Square Error (RMSE) values for the best models for Random Forest ranges from 4.74 to 6.37. The R\textsuperscript{2} values for the best models for Random Forest ranges from .91 to .96. The accuracy achieved using RMSE values for the best models for KNN ranges from 5.00 to 7.72. The R\textsuperscript{2} values for the best models for KNN ranges from .89 to .95. The best imputation models received R\textsuperscript{2} between .89 to .96 and RMSE values less than 8dB. We also show that the accuracy of classification predictive models performed better with our best imputation models versus constant imputations by a two percent increase.

Keywords: machine learning, audiograms, data imputations, single imputations

Procedia PDF Downloads 52

Authors: Sondos M. Hammad, Sherin M. ElGokhy, Mahmoud M. Fahmy, Elsayed A. Sallam

Abstract:

In this paper, an ensemble-based Support Vector Machine (SVM) classification approach is proposed. It is used for miRNA prediction. Three problems, commonly associated with previous approaches, are alleviated. These problems arise due to impose assumptions on the secondary structural of premiRNA, imbalance between the numbers of the laboratory checked miRNAs and the pseudo-hairpins, and finally using a training data set that does not consider all the varieties of samples in different species. We aggregate the predicted outputs of three well-known SVM classifiers; namely, Triplet-SVM, Virgo and Mirident, weighted by their variant features without any structural assumptions. An additional SVM layer is used in aggregating the final output. The proposed approach is trained and then tested with balanced data sets. The results of the proposed approach outperform the three base classifiers. Improved values for the metrics of 88.88% f-score, 92.73% accuracy, 90.64% precision, 96.64% specificity, 87.2% sensitivity, and the area under the ROC curve is 0.91 are achieved.

Keywords: MiRNAs, SVM classification, ensemble algorithm, assumption problem, imbalance data

Procedia PDF Downloads 307

Authors: Kaewkarn Phuangsombat, Arthit Phuangsombat, Anupun Terdwongworakul

Abstract:

Hard seeds will not grow and can cause mold in sprouting process. Thus, the hard seeds need to be separated from the normal seeds. Near infrared hyperspectral imaging in a range of 900 to 1700 nm was implemented to develop a model by partial least squares discriminant analysis to discriminate the hard seeds from the normal seeds. The orientation of the seeds was also studied to compare the performance of the models. The model based on hilum-up orientation achieved the best result giving the coefficient of determination of 0.98, and root mean square error of prediction of 0.07 with classification accuracy was equal to 100%.

Keywords: mung bean, near infrared, germinatability, hard seed

Procedia PDF Downloads 273

Authors: Rosa L. Figueroa, Christopher A. Flores, Qing Zeng-Treitler

Abstract:

A regular expression consists of sequence characters which allow describing a text path. Usually, in clinical research, regular expressions are manually created by programmers together with domain experts. Lately, there have been several efforts to investigate how to generate them automatically. This article presents a text classification algorithm based on regexes. The algorithm named REX was designed, and then, implemented as a simplified method to create regexes to classify Spanish text automatically. In order to classify ambiguous cases, such as, when multiple labels are assigned to a testing example, REX includes an information gain method Two sets of data were used to evaluate the algorithm’s effectiveness in clinical text classification tasks. The results indicate that the regular expression based classifier proposed in this work performs statically better regarding accuracy and F-measure than Support Vector Machine and Naïve Bayes for both datasets.

Keywords: information gain, regular expressions, smith-waterman algorithm, text classification

Procedia PDF Downloads 292

Authors: Shelby Simpson, William Stanley, Namir Naba, Xiaodi Wang

Abstract:

Both t-SNE and UMAP are brand new state of art tools to predominantly preserve the local structure that is to group neighboring data points together, which indeed provides a very informative visualization of heterogeneity in our data. In this research, we develop a t-SNE and UMAP base neural network image classification algorithm to embed the original dataset to a corresponding low dimensional dataset as a preprocessing step, then use this embedded database as input to our specially designed neural network classifier for image classification. We use the fashion MNIST data set, which is a labeled data set of images of clothing objects in our experiments. t-SNE and UMAP are used for dimensionality reduction of the data set and thus produce low dimensional embeddings. Furthermore, we use the embeddings from t-SNE and UMAP to feed into two neural networks. The accuracy of the models from the two neural networks is then compared to a dense neural network that does not use embedding as an input to show which model can classify the images of clothing objects more accurately.

Keywords: t-SNE, UMAP, fashion MNIST, neural networks

Procedia PDF Downloads 164

Authors: Bingchun Liu, Pei-Chann Chang, Natasha Huang, Dun Li

Abstract:

Machine Learning and Data Mining are the two important tools for extracting useful information and knowledge from large datasets. In machine learning, classification is a wildly used technique to predict qualitative variables and is generally preferred over regression from an operational point of view. Due to the enormous increase in air pollution in various countries especially China, Air Quality Classification has become one of the most important topics in air quality research and modelling. This study aims at introducing a hybrid classification model based on information theory and Support Vector Machine (SVM) using the air quality data of four cities in China namely Beijing, Guangzhou, Shanghai and Tianjin from Jan 1, 2014 to April 30, 2016. China's Ministry of Environmental Protection has classified the daily air quality into 6 levels namely Serious Pollution, Severe Pollution, Moderate Pollution, Light Pollution, Good and Excellent based on their respective Air Quality Index (AQI) values. Using the information theory, information gain (IG) is calculated and feature selection is done for both categorical features and continuous numeric features. Then SVM Machine Learning algorithm is implemented on the selected features with cross-validation. The final evaluation reveals that the IG and SVM hybrid model performs better than SVM (alone), Artificial Neural Network (ANN) and K-Nearest Neighbours (KNN) models in terms of accuracy as well as complexity.

Keywords: machine learning, air quality classification, air quality index, information gain, support vector machine, cross-validation

Procedia PDF Downloads 199

Authors: Mekamı Hayet, Bounoua Nacer, Benabderrahmane Sidahmed, Taleb Ahmed

Abstract:

Pose estimation is an important task in computer vision. Though the majority of the existing solutions provide good accuracy results, they are often overly complex and computationally expensive. In this perspective, we propose the use of dimensionality reduction techniques to address the problem of facial pose estimation. Firstly, a face image is converted into one-dimensional time series using Hilbert space filling curve, then the approach converts these time series data to a symbolic representation. Furthermore, a distance matrix is calculated between symbolic series of an input learning dataset of images, to generate classifiers of frontal vs. profile face pose. The proposed method is evaluated with three public datasets. Experimental results have shown that our approach is able to achieve a correct classification rate exceeding 97% with K-NN algorithm.

Keywords: machine learning, pattern recognition, facial pose classification, time series

Procedia PDF Downloads 321

Authors: Filippo Portera

Abstract:

Usually standard algorithms employ a loss where each error is the mere absolute difference between the true value and the prediction, in case of a regression task. In the present, we present several error weighting schemes that are a generalization of the consolidated routine. We study both a binary classification model for Support Vextor Classification and a regression net for Multylayer Perceptron. Results proves that the error is never worse than the standard procedure and several times it is better.

Keywords: loss, binary-classification, MLP, weights, regression

Procedia PDF Downloads 63

Authors: Moschos Vogiatzis, K. Perakis

Abstract:

Spatial-explicit and up-to-date land use/land cover information is fundamental for spatial planning, land management, sustainable development, and sound decision-making. In the last decade, many satellite-derived land cover products at different spatial, spectral, and temporal resolutions have been developed, such as the European Copernicus Land Cover product. However, more efficient and detailed information for land use/land cover is required at the regional or local scale. A typical Mediterranean basin with a complex landscape comprised of various forest types, crops, artificial surfaces, and wetlands was selected to test and develop our approach. In this study, we investigate the improvement of Copernicus Land Cover product (CLC2018) using Landsat 8 and Sentinel-2 pixel-based classification based on all available existing geospatial data (Forest Maps, LPIS, Natura2000 habitats, cadastral parcels, etc.). We examined and compared the performance of the Random Forest classifier for land use/land cover mapping. In total, 10 land use/land cover categories were recognized in Landsat 8 and 11 in Sentinel-2A. A comparison of the overall classification accuracies for 2018 shows that Landsat 8 classification accuracy was slightly higher than Sentinel-2A (82,99% vs. 80,30%). We concluded that the main land use/land cover types of CLC2018, even within a heterogeneous area, can be successfully mapped and updated according to CLC nomenclature. Future research should be oriented toward integrating spatiotemporal information from seasonal bands and spectral indexes in the classification process.

Keywords: classification, land use/land cover, mapping, random forest

Procedia PDF Downloads 98

Authors: Bilel Chalghaf, Mathieu Varin

Abstract:

Forest characterization in Quebec, Canada, is usually assessed based on photo-interpretation at the stand level. For species identification, this often results in a lack of precision. Very high spatial resolution imagery, such as DigitalGlobe, and Light Detection and Ranging (LiDAR), have the potential to overcome the limitations of aerial imagery. To date, few studies have used that data to map a large number of species at the tree level using machine learning techniques. The main objective of this study is to map 11 individual high tree species ( > 17m) at the tree level using an object-based approach in the broadleaf forest of Kenauk Nature, Quebec. For the individual tree crown segmentation, three canopy-height models (CHMs) from LiDAR data were assessed: 1) the original, 2) a filtered, and 3) a corrected model. The corrected CHM gave the best accuracy and was then coupled with imagery to refine tree species crown identification. When compared with photo-interpretation, 90% of the objects represented a single species. For modeling, 313 variables were derived from 16-band WorldView-3 imagery and LiDAR data, using radiance, reflectance, pixel, and object-based calculation techniques. Variable selection procedures were employed to reduce their number from 313 to 16, using only 11 bands to aid reproducibility. For classification, a global approach using all 11 species was compared to a semi-hierarchical hybrid classification approach at two levels: (1) tree type (broadleaf/conifer) and (2) individual broadleaf (five) and conifer (six) species. Five different model techniques were used: (1) support vector machine (SVM), (2) classification and regression tree (CART), (3) random forest (RF), (4) k-nearest neighbors (k-NN), and (5) linear discriminant analysis (LDA). Each model was tuned separately for all approaches and levels. For the global approach, the best model was the SVM using eight variables (overall accuracy (OA): 80%, Kappa: 0.77). With the semi-hierarchical hybrid approach, at the tree type level, the best model was the k-NN using six variables (OA: 100% and Kappa: 1.00). At the level of identifying broadleaf and conifer species, the best model was the SVM, with OA of 80% and 97% and Kappa values of 0.74 and 0.97, respectively, using seven variables for both models. This paper demonstrates that a hybrid classification approach gives better results and that using 16-band WorldView-3 with LiDAR data leads to more precise predictions for tree segmentation and classification, especially when the number of tree species is large.

Keywords: tree species, object-based, classification, multispectral, machine learning, WorldView-3, LiDAR

Procedia PDF Downloads 106

Authors: Salima Smiti, Ines Gasmi, Makram Soui

Abstract:

Credit risk is the most important issue for financial institutions. Its assessment becomes an important task used to predict defaulter customers and classify customers as good or bad payers. To this objective, numerous techniques have been applied for credit risk assessment. However, to our knowledge, several evaluation techniques are black-box models such as neural networks, SVM, etc. They generate applicants’ classes without any explanation. In this paper, we propose to assess credit risk using rules classification method. Our output is a set of rules which describe and explain the decision. To this end, we will compare seven classification algorithms (JRip, Decision Table, OneR, ZeroR, Fuzzy Rule, PART and Genetic programming (GP)) where the goal is to find the best rules satisfying many criteria: accuracy, sensitivity, and specificity. The obtained results confirm the efficiency of the GP algorithm for German and Australian datasets compared to other rule-based techniques to predict the credit risk.

Keywords: credit risk assessment, classification algorithms, data mining, rule extraction

Procedia PDF Downloads 145

Authors: Zachary Pan

Abstract:

Our general mission is to solve the problem of classifying images into different body part types and deciding if each of them is healthy or not. However, for now, we will determine healthiness for only one-sixth of the body parts, specifically the chest. We will detect pneumonia in X-ray scans of those chest images. With this type of AI, doctors can use it as a second opinion when they are taking CT or X-ray scans of their patients. Another ad-vantage of using this machine learning classifier is that it has no human weaknesses like fatigue. The overall ap-proach to this problem is to split the problem into two parts: first, classify the image, then determine if it is healthy. In order to classify the image into a specific body part class, the body parts dataset must be split into test and training sets. We can then use many models, like neural networks or logistic regression models, and fit them using the training set. Now, using the test set, we can obtain a realistic accuracy the models will have on images in the real world since these testing images have never been seen by the models before. In order to increase this testing accuracy, we can also apply many complex algorithms to the models, like multiplicative weight update. For the second part of the problem, to determine if the body part is healthy, we can have another dataset consisting of healthy and non-healthy images of the specific body part and once again split that into the test and training sets. We then use another neural network to train on those training set images and use the testing set to figure out its accuracy. We will do this process only for the chest images. A major conclusion reached is that convolutional neural networks are the most reliable and accurate at image classification. In classifying the images, the logistic regression model, the neural network, neural networks with multiplicative weight update, neural networks with the black box algorithm, and the convolutional neural network achieved 96.83 percent accuracy, 97.33 percent accuracy, 97.83 percent accuracy, 96.67 percent accuracy, and 98.83 percent accuracy, respectively. On the other hand, the overall accuracy of the model that de-termines if the images are healthy or not is around 78.37 percent accuracy.

Keywords: body part, healthcare, machine learning, neural networks

Procedia PDF Downloads 71

Authors: Yunwei Tang, Linhai Jing, Hui Li, Qingjie Liu, Xiuxia Li, Qi Yan, Haifeng Ding

Abstract:

The giant panda (Ailuropoda melanoleuca) is an endangered species, mainly live in central China, where bamboos act as the main food source of wild giant pandas. Knowledge of spatial distribution of bamboos therefore becomes important for identifying the habitat of giant pandas. There have been ongoing studies for mapping bamboos and other tree species using remote sensing. WorldView-2 (WV-2) is the first high resolution commercial satellite with eight Multi-Spectral (MS) bands. Recent studies demonstrated that WV-2 imagery has a high potential in classification of tree species. The advanced classification techniques are important for utilising high spatial resolution imagery. It is generally agreed that object-based image analysis is a more desirable method than pixel-based analysis in processing high spatial resolution remotely sensed data. Classifiers that use spatial information combined with spectral information are known as contextual classifiers. It is suggested that contextual classifiers can achieve greater accuracy than non-contextual classifiers. Thus, spatial correlation can be incorporated into classifiers to improve classification results. The study area is located at Wuyipeng area in Wolong, Sichuan Province. The complex environment makes it difficult for information extraction since bamboos are sparsely distributed, mixed with brushes, and covered by other trees. Extensive fieldworks in Wuyingpeng were carried out twice. The first one was on 11th June, 2014, aiming at sampling feature locations for geometric correction and collecting training samples for classification. The second fieldwork was on 11th September, 2014, for the purposes of testing the classification results. In this study, spectral separability analysis was first performed to select appropriate MS bands for classification. Also, the reflectance analysis provided information for expanding sample points under the circumstance of knowing only a few. Then, a spatially weighted object-based k-nearest neighbour (k-NN) classifier was applied to the selected MS bands to identify seven land cover types (bamboo, conifer, broadleaf, mixed forest, brush, bare land, and shadow), accounting for spatial correlation within classes using geostatistical modelling. The spatially weighted k-NN method was compared with three alternatives: the traditional k-NN classifier, the Support Vector Machine (SVM) method and the Classification and Regression Tree (CART). Through field validation, it was proved that the classification result obtained using the spatially weighted k-NN method has the highest overall classification accuracy (77.61%) and Kappa coefficient (0.729); the producer’s accuracy and user’s accuracy achieve 81.25% and 95.12% for the bamboo class, respectively, also higher than the other methods. Photos of tree crowns were taken at sample locations using a fisheye camera, so the canopy density could be estimated. It is found that it is difficult to identify bamboo in the areas with a large canopy density (over 0.70); it is possible to extract bamboos in the areas with a median canopy density (from 0.2 to 0.7) and in a sparse forest (canopy density is less than 0.2). In summary, this study explores the ability of WV-2 imagery for bamboo extraction in a mountainous region in Sichuan. The study successfully identified the bamboo distribution, providing supporting knowledge for assessing the habitats of giant pandas.

Keywords: bamboo mapping, classification, geostatistics, k-NN, worldview-2

Procedia PDF Downloads 289

Authors: Eugene Y. J. Aw, J. W. Koh, S. H. Chew, K. E. Chua, Lionel L. J. Ang, Algernon C. S. Hong, Danette S. E. Tan, Grace H. B. Foo, K. Q. Hong, L. M. Cheng, M. L. Leong

Abstract:

This paper presents a novel rapid soil classification technique that combines computer vision with four-probe soil electrical resistivity method and cone penetration test (CPT), to improve the accuracy and productivity of on-site classification of excavated soil. In Singapore, excavated soils from local construction projects are transported to Staging Grounds (SGs) to be reused as fill material for land reclamation. Excavated soils are mainly categorized into two groups (“Good Earth” and “Soft Clay”) based on particle size distribution (PSD) and water content (w) from soil investigation reports and on-site visual survey, such that proper treatment and usage can be exercised. However, this process is time-consuming and labour-intensive. Thus, a rapid classification method is needed at the SGs. Computer vision, four-probe soil electrical resistivity and CPT were combined into an innovative non-destructive and instantaneous classification method for this purpose. The computer vision technique comprises soil image acquisition using industrial grade camera; image processing and analysis via calculation of Grey Level Co-occurrence Matrix (GLCM) textural parameters; and decision-making using an Artificial Neural Network (ANN). Complementing the computer vision technique, the apparent electrical resistivity of soil (ρ) is measured using a set of four probes arranged in Wenner’s array. It was found from the previous study that the ANN model coupled with ρ can classify soils into “Good Earth” and “Soft Clay” in less than a minute, with an accuracy of 85% based on selected representative soil images. To further improve the technique, the soil strength is measured using a modified mini cone penetrometer, and w is measured using a set of time-domain reflectometry (TDR) probes. Laboratory proof-of-concept was conducted through a series of seven tests with three types of soils – “Good Earth”, “Soft Clay” and an even mix of the two. Validation was performed against the PSD and w of each soil type obtained from conventional laboratory tests. The results show that ρ, w and CPT measurements can be collectively analyzed to classify soils into “Good Earth” or “Soft Clay”. It is also found that these parameters can be integrated with the computer vision technique on-site to complete the rapid soil classification in less than three minutes.

Keywords: Computer vision technique, cone penetration test, electrical resistivity, rapid and non-destructive, soil classification

Procedia PDF Downloads 178

Authors: Gulfam Haider, sana danish

Abstract:

Convolutional Neural Networks (CNNs) have emerged as powerful tools for image classification, and the choice of optimizers profoundly affects their performance. The study of optimizers and their adaptations remains a topic of significant importance in machine learning research. While numerous studies have explored and advocated for various optimizers, the efficacy of these optimization techniques is still subject to scrutiny. This work aims to address the challenges surrounding the effectiveness of optimizers by conducting a comprehensive analysis and evaluation. The primary focus of this investigation lies in examining the performance of different optimizers when employed in conjunction with the popular activation function, Rectified Linear Unit (ReLU). By incorporating ReLU, known for its favorable properties in prior research, the aim is to bolster the effectiveness of the optimizers under scrutiny. Specifically, we evaluate the adjustment of these optimizers with both the original Softmax activation function and the modified ReLU activation function, carefully assessing their impact on overall performance. To achieve this, a series of experiments are conducted using a well-established benchmark dataset for image classification tasks, namely the Canadian Institute for Advanced Research dataset (CIFAR-10). The selected optimizers for investigation encompass a range of prominent algorithms, including Adam, Root Mean Squared Propagation (RMSprop), Adaptive Learning Rate Method (Adadelta), Adaptive Gradient Algorithm (Adagrad), and Stochastic Gradient Descent (SGD). The performance analysis encompasses a comprehensive evaluation of the classification accuracy, convergence speed, and robustness of the CNN models trained with each optimizer. Through rigorous experimentation and meticulous assessment, we discern the strengths and weaknesses of the different optimization techniques, providing valuable insights into their suitability for image classification tasks. By conducting this in-depth study, we contribute to the existing body of knowledge surrounding optimizers in CNNs, shedding light on their performance characteristics for image classification. The findings gleaned from this research serve to guide researchers and practitioners in making informed decisions when selecting optimizers and activation functions, thus advancing the state-of-the-art in the field of image classification with convolutional neural networks.

Keywords: deep neural network, optimizers, RMsprop, ReLU, stochastic gradient descent

Procedia PDF Downloads 60

Authors: Mayank Gupta, Siba Prasad Samal, Vasu Kakkirala

Abstract:

The internet world and its priorities have changed considerably in the last decade. Browsing on smart phones has increased manifold and is set to explode much more. Users spent considerable time browsing different websites, that gives a great deal of insight into user’s preferences. Instead of plain information classifying different aspects of browsing like Bookmarks, History, and Download Manager into useful categories would improve and enhance the user’s experience. Most of the classification solutions are server side that involves maintaining server and other heavy resources. It has security constraints and maybe misses on contextual data during classification. On device, classification solves many such problems, but the challenge is to achieve accuracy on classification with resource constraints. This on device classification can be much more useful in personalization, reducing dependency on cloud connectivity and better privacy/security. This approach provides more relevant results as compared to current standalone solutions because it uses content rendered by browser which is customized by the content provider based on user’s profile. This paper proposes a Naive Bayes based lightweight classification engine targeted for a resource constraint devices. Our solution integrates with Web Browser that in turn triggers classification algorithm. Whenever a user browses a webpage, this solution extracts DOM Tree data from the browser’s rendering engine. This DOM data is a dynamic, contextual and secure data that can’t be replicated. This proposal extracts different features of the webpage that runs on an algorithm to classify into multiple categories. Naive Bayes based engine is chosen in this solution for its inherent advantages in using limited resources compared to other classification algorithms like Support Vector Machine, Neural Networks, etc. Naive Bayes classification requires small memory footprint and less computation suitable for smartphone environment. This solution has a feature to partition the model into multiple chunks that in turn will facilitate less usage of memory instead of loading a complete model. Classification of the webpages done through integrated engine is faster, more relevant and energy efficient than other standalone on device solution. This classification engine has been tested on Samsung Z3 Tizen hardware. The Engine is integrated into Tizen Browser that uses Chromium Rendering Engine. For this solution, extensive dataset is sourced from dmoztools.net and cleaned. This cleaned dataset has 227.5K webpages which are divided into 8 generic categories ('education', 'games', 'health', 'entertainment', 'news', 'shopping', 'sports', 'travel'). Our browser integrated solution has resulted in 15% less memory usage (due to partition method) and 24% less power consumption in comparison with standalone solution. This solution considered 70% of the dataset for training the data model and the rest 30% dataset for testing. An average accuracy of ~96.3% is achieved across the above mentioned 8 categories. This engine can be further extended for suggesting Dynamic tags and using the classification for differential uses cases to enhance browsing experience.

Keywords: chromium, lightweight engine, mobile computing, Naive Bayes, Tizen, web browser, webpage classification

Procedia PDF Downloads 136

Authors: Jaishree Ranganathan, MuthuPriya Shanmugakani Velsamy, Shamika Kulkarni, Angelina Tzacheva

Abstract:

Emotions play an important role in everyday life. An-alyzing these emotions or feelings from social media platforms like Twitter, Facebook, blogs, and forums based on user comments and reviews plays an important role in various factors. Some of them include brand monitoring, marketing strategies, reputation, and competitor analysis. The opinions or sentiments mined from such data helps understand the current state of the user. It does not directly provide intuitive insights on what actions to be taken to benefit the end user or business. Actionable Pattern Mining method provides suggestions or actionable recommendations on what changes or actions need to be taken in order to benefit the end user. In this paper, we propose automatic classification of emotions in Twitter data using Recurrent Neural Network - Gated Recurrent Unit. We achieve training accuracy of 87.58% and validation accuracy of 86.16%. Also, we extract action rules with respect to the user emotion that helps to provide actionable suggestion.

Keywords: emotion mining, twitter, recurrent neural network, gated recurrent unit, actionable pattern mining

Procedia PDF Downloads 133

Authors: Nicholas Milburn, Yu Liang, Dalei Wu

Abstract:

Athlete performance scoring within the climbing do-main presents interesting challenges as the sport does not have an objective way to assign skill. Assessing skill levels within any sport is valuable as it can be used to mark progress while training, and it can help an athlete choose appropriate climbs to attempt. Machine learning-based methods are popular for complex problems like this. The dataset available was composed of dynamic force data recorded during climbing; however, this dataset came with challenges such as data scarcity, imbalance, and it was temporally heterogeneous. Investigated solutions to these challenges include data augmentation, temporal normalization, conversion of time series to the spectral domain, and cross validation strategies. The investigated solutions to the classification problem included light weight machine classifiers KNN and SVM as well as the deep learning with CNN. The best performing model had an 80% accuracy. In conclusion, there seems to be enough information within climbing force data to accurately categorize climbers by skill.

Keywords: classification, climbing, data imbalance, data scarcity, machine learning, time sequence

Procedia PDF Downloads 116

Authors: Shourav Chowdhury, Subrata Barua, K. M. Naimuddin, Imam Hassan Sajib, Md. Hasan, Shudipta Banik, Muna Das

Abstract:

Traffic accidents stand as a leading cause of fatalities worldwide, significantly impacting global mortality rates. Accurate classification of road accidents through advanced technological solutions presents a crucial opportunity to revolutionize accident prevention and emergency response strategies. This paper presents an advanced deep-learning methodology customized for the classification of road accidents using CCTV surveillance footage. This real-time dataset, comprising approximately 18,000 frames, has been amassed, which is pivotal for enabling comprehensive research in this field. This substantial dataset is the foundation for these investigative efforts, providing a rich and diverse source for conducting an in-depth analysis of the features. It has achieved a remarkable accuracy of 97% on this dataset through the strategic utilization of transfer learning in conjunction with LSTM (Long short-term memory) techniques. This accomplishment underscores the efficacy of our approach, combining the strengths of transfer learning and LSTM models, resulting in a highly accurate classification system for road accident events.

Keywords: accident, CCTV, footage, long short-term memory, surveillance

Procedia PDF Downloads 21

Authors: Po-Jen Chen, Jian-Jiun Ding, Hung-Wei Hsu, Chien-Yao Wang, Jia-Ching Wang

Abstract:

A more accurate scene classification algorithm using label relations and the hierarchy neural network was developed in this work. In many classification algorithms, it is assumed that the labels are mutually exclusive. This assumption is true in some specific problems, however, for scene classification, the assumption is not reasonable. Because there are a variety of objects with a photo image, it is more practical to assign multiple labels for an image. In this paper, two label relations, which are exclusive relation and hierarchical relation, were adopted in the classification process to achieve more accurate multiple label classification results. Moreover, the hierarchy neural network (hierarchy NN) is applied to classify the image and the directed acyclic graph structure is used for predicting a more reasonable result which obey exclusive and hierarchical relations. Simulations show that, with these techniques, a much more accurate scene classification result can be achieved.

Keywords: convolutional neural network, label relation, hierarchy neural network, scene classification

Procedia PDF Downloads 423

Authors: Thiago Spilborghs Bueno Meyer, Plinio Thomaz Aquino Junior

Abstract:

Through speech, which privileges the functional and interactive nature of the text, it is possible to ascertain the spatiotemporal circumstances, the conditions of production and reception of the discourse, the explicit purposes such as informing, explaining, convincing, etc. These conditions allow bringing the interaction between humans closer to the human-robot interaction, making it natural and sensitive to information. However, it is not enough to understand what is said; it is necessary to recognize emotions for the desired interaction. The validity of the use of neural networks for feature selection and emotion recognition was verified. For this purpose, it is proposed the use of neural networks and comparison of models, such as recurrent neural networks and deep neural networks, in order to carry out the classification of emotions through speech signals to verify the quality of recognition. It is expected to enable the implementation of robots in a domestic environment, such as the HERA robot from the RoboFEI@Home team, which focuses on autonomous service robots for the domestic environment. Tests were performed using only the Mel-Frequency Cepstral Coefficients, as well as tests with several characteristics of Delta-MFCC, spectral contrast, and the Mel spectrogram. To carry out the training, validation and testing of the neural networks, the eNTERFACE’05 database was used, which has 42 speakers from 14 different nationalities speaking the English language. The data from the chosen database are videos that, for use in neural networks, were converted into audios. It was found as a result, a classification of 51,969% of correct answers when using the deep neural network, when the use of the recurrent neural network was verified, with the classification with accuracy equal to 44.09%. The results are more accurate when only the Mel-Frequency Cepstral Coefficients are used for the classification, using the classifier with the deep neural network, and in only one case, it is possible to observe a greater accuracy by the recurrent neural network, which occurs in the use of various features and setting 73 for batch size and 100 training epochs.

Keywords: emotion recognition, speech, deep learning, human-robot interaction, neural networks

Procedia PDF Downloads 130

Authors: Ahmed ElAzab, Amira M. Idrees, Mahmoud A. Mahmoud, Hesham Hefny

Abstract:

Social networking sites such as Twitter and Facebook attracts over 500 million users across the world, for those users, their social life, even their practical life, has become interrelated. Their interaction with social networking has affected their life forever. Accordingly, social networking sites have become among the main channels that are responsible for vast dissemination of different kinds of information during real time events. This popularity in Social networking has led to different problems including the possibility of exposing incorrect information to their users through fake accounts which results to the spread of malicious content during life events. This situation can result to a huge damage in the real world to the society in general including citizens, business entities, and others. In this paper, we present a classification method for detecting fake accounts on Twitter. The study determines the minimized set of the main factors that influence the detection of the fake accounts on Twitter, then the determined factors have been applied using different classification techniques, a comparison of the results for these techniques has been performed and the most accurate algorithm is selected according to the accuracy of the results. The study has been compared with different recent research in the same area, this comparison has proved the accuracy of the proposed study. We claim that this study can be continuously applied on Twitter social network to automatically detect the fake accounts, moreover, the study can be applied on different Social network sites such as Facebook with minor changes according to the nature of the social network which are discussed in this paper.

Keywords: fake accounts detection, classification algorithms, twitter accounts analysis, features based techniques

Procedia PDF Downloads 368

Authors: Yingisani Chabalala, Elhadi Adam, Khalid Adem Ali

Abstract:

Mapping smallholder fruit plantations using optical data is challenging due to morphological landscape heterogeneity and crop types having overlapped spectral signatures. Furthermore, cloud covers limit the use of optical sensing, especially in subtropical climates where they are persistent. This research assessed the effectiveness of Sentinel-1 (S1) and Sentinel-2 (S2) data for mapping fruit trees and co-existing land-use types by using support vector machine (SVM) and random forest (RF) classifiers independently. These classifiers were also applied to fused data from the two sensors. Feature ranks were extracted using the RF mean decrease accuracy (MDA) and forward variable selection (FVS) to identify optimal spectral windows to classify fruit trees. Based on RF MDA and FVS, the SVM classifier resulted in relatively high classification accuracy with overall accuracy (OA) = 0.91.6% and kappa coefficient = 0.91% when applied to the fused satellite data. Application of SVM to S1, S2, S2 selected variables and S1S2 fusion independently produced OA = 27.64, Kappa coefficient = 0.13%; OA= 87%, Kappa coefficient = 86.89%; OA = 69.33, Kappa coefficient = 69. %; OA = 87.01%, Kappa coefficient = 87%, respectively. Results also indicated that the optimal spectral bands for fruit tree mapping are green (B3) and SWIR_2 (B10) for S2, whereas for S1, the vertical-horizontal (VH) polarization band. Including the textural metrics from the VV channel improved crop discrimination and co-existing land use cover types. The fusion approach proved robust and well-suited for accurate smallholder fruit plantation mapping.

Keywords: smallholder agriculture, fruit trees, data fusion, precision agriculture

Procedia PDF Downloads 19

Authors: Nadia el Atlas, Mohammed el Aroussi, Mohammed Wahbi

Abstract:

In this paper a novel technique of mass characterization based on robust features-fusion is presented. The proposed method consists of mainly four stages: (a) the first phase involves segmenting the masses using edge information’s. (b) The second phase is to calculate and fuse the most relevant morphological features. (c) The last phase is the classification step which allows us to classify the images into benign and malignant masses. In this step we have implemented Support Vectors Machines (SVM) and Artificial Neural Networks (ANN), which were evaluated with the following performance criteria: confusion matrix, accuracy, sensitivity, specificity, receiver operating characteristic ROC, and error histogram. The effectiveness of this new approach was evaluated by a recently developed database: INBREAST database. The fusion of the most appropriate morphological features provided very good results. The SVM gives accuracy to within 64.3%. Whereas the ANN classifier gives better results with an accuracy of 97.5%.

Keywords: breast cancer, mammography, CAD system, features, fusion

Procedia PDF Downloads 566

Authors: Sanchali Das, Swapan Debbarma

Abstract:

Music mood classification is developing in both the areas of music information retrieval (MIR) and natural language processing (NLP). Some of the Indian languages like Hindi English etc. have considerable exposure in MIR. But research in mood classification in regional language is very less. In this paper, powerful audio based feature for Kokborok Christian song is identified and mood classification task has been performed. Kokborok is an Indo-Burman language especially spoken in the northeastern part of India and also some other countries like Bangladesh, Myanmar etc. For performing audio-based classification task, useful audio features are taken out by jMIR software. There are some standard audio parameters are there for the audio-based task but as known to all that every language has its unique characteristics. So here, the most significant features which are the best fit for the database of Kokborok song is analysed. The regression-based model is used to find out the independent parameters that act as a predictor and predicts the dependencies of parameters and shows how it will impact on overall classification result. For classification WEKA 3.5 is used, and selected parameters create a classification model. And another model is developed by using all the standard audio features that are used by most of the researcher. In this experiment, the essential parameters that are responsible for effective audio based mood classification and parameters that do not significantly change for each of the Christian Kokborok songs are analysed, and a comparison is also shown between the two above model.

Keywords: Christian Kokborok song, mood classification, music information retrieval, regression

Procedia PDF Downloads 191

Authors: Aishrith Rao

Abstract:

COVID 19 is a pandemic which has resulted in thousands of deaths around the world and a huge impact on the global economy. Testing is a huge issue as the test kits have limited availability and are expensive to manufacture. Using deep learning methods on radiology images in the detection of the coronavirus as these images contain information about the spread of the virus in the lungs is extremely economical and time-saving as it can be used in areas with a lack of testing facilities. This paper focuses on binary classification and multi-class classification of COVID 19 and other diseases such as pneumonia, tuberculosis, etc. Different deep learning methods such as VGG-19, COVID-Net, ResNET+ SVM, Deep CNN, DarkCovidnet, etc., have been used, and their accuracy has been compared using the Chest X-Ray dataset.

Keywords: deep learning, computer vision, radiology, COVID-19, ResNet, VGG-19, deep neural networks

Procedia PDF Downloads 118

Authors: Cuneyt Yucelbas, Seral Ozsen, Sule Yucelbas, Gulay Tezel

Abstract:

Due to the fact that there exist only a small number of complex systems in artificial immune system (AIS) that work out nonlinear problems, nonlinear AIS approaches, among the well-known solution techniques, need to be developed. Gaussian function is usually used as similarity estimation in classification problems and pattern recognition. In this study, diagnosis of breast cancer, the second type of the most widespread cancer in women, was performed with different distance calculation functions that euclidean, gaussian and gaussian-euclidean hybrid function in the clonal selection model of classical AIS on Wisconsin Breast Cancer Dataset (WBCD), which was taken from the University of California, Irvine Machine-Learning Repository. We used 3-fold cross validation method to train and test the dataset. According to the results, the maximum test classification accuracy was reported as 97.35% by using of gaussian-euclidean hybrid function for fold-3. Also, mean of test classification accuracies for all of functions were obtained as 94.78%, 94.45% and 95.31% with use of euclidean, gaussian and gaussian-euclidean, respectively. With these results, gaussian-euclidean hybrid function seems to be a potential distance calculation method, and it may be considered as an alternative distance calculation method for hard nonlinear classification problems.

Keywords: artificial immune system, breast cancer diagnosis, Euclidean function, Gaussian function

Procedia PDF Downloads 412

Authors: Muneeb Ullah, Daishihan, Xiadong Young

Abstract:

Deep learning plays a crucial role in identifying COVID-19 and preventing its spread. To improve the accuracy of COVID-19 diagnoses, it is important to have access to a sufficient number of training images of CXRs (chest X-rays) depicting the disease. However, there is currently a shortage of such images. To address this issue, this paper introduces COVID-19 GAN, a model that uses generative adversarial networks (GANs) to generate realistic CXR images of COVID-19, which can be used to train identification models. Initially, a generator model is created that uses digressive channels to generate images of CXR scans for COVID-19. To differentiate between real and fake disease images, an efficient discriminator is developed by combining the dense connectivity strategy and instance normalization. This approach makes use of their feature extraction capabilities on CXR hazy areas. Lastly, the deep regret gradient penalty technique is utilized to ensure stable training of the model. With the use of 4,062 grape leaf disease images, the Leaf GAN model successfully produces 8,124 COVID-19 CXR images. The COVID-19 GAN model produces COVID-19 CXR images that outperform DCGAN and WGAN in terms of the Fréchet inception distance. Experimental findings suggest that the COVID-19 GAN-generated CXR images possess noticeable haziness, offering a promising approach to address the limited training data available for COVID-19 model training. When the dataset was expanded, CNN-based classification models outperformed other models, yielding higher accuracy rates than those of the initial dataset and other augmentation techniques. Among these models, ImagNet exhibited the best recognition accuracy of 99.70% on the testing set. These findings suggest that the proposed augmentation method is a solution to address overfitting issues in disease identification and can enhance identification accuracy effectively.

Keywords: classification, deep learning, medical images, CXR, GAN.

Procedia PDF Downloads 52

Authors: Marielcio Lacerda, Angelo Paulino, Elcio Shiguemori, Alvaro Damiao, Lamartine Guimaraes, Camila Anjos

Abstract:

Accurate information about the terrain is extremely important in disaster management activities or conflict. This paper proposes the use of the Unmanned Aircraft Systems (UAS) at the identification of Airdrop Zones (AZs) and Helicopter Landing Zones (HLZs). In this paper we consider the AZs the zones where troops or supplies are dropped by parachute, and HLZs areas where victims can be rescued. The use of digital image processing enables the automatic generation of an orthorectified mosaic and an actual Digital Surface Model (DSM). This methodology allows obtaining this fundamental information to the terrain’s comprehension post-disaster in a short amount of time and with good accuracy. In order to get the identification and classification of AZs and HLZs images from DJI drone, model Phantom 4 have been used. The images were obtained with the knowledge and authorization of the responsible sectors and were duly registered in the control agencies. The flight was performed on May 24, 2017, and approximately 1,300 images were obtained during approximately 1 hour of flight. Afterward, new attributes were generated by Feature Extraction (FE) from the original images. The use of multispectral images and complementary attributes generated independently from them increases the accuracy of classification. The attributes of this work include the Declivity Map and Principal Component Analysis (PCA). For the classification four distinct classes were considered: HLZ 1 – small size (18m x 18m); HLZ 2 – medium size (23m x 23m); HLZ 3 – large size (28m x 28m); AZ (100m x 100m). The Decision Tree method Random Forest (RF) was used in this work. RF is a classification method that uses a large collection of de-correlated decision trees. Different random sets of samples are used as sampled objects. The results of classification from each tree and for each object is called a class vote. The resulting classification is decided by a majority of class votes. In this case, we used 200 trees for the execution of RF in the software WEKA 3.8. The classification result was visualized on QGIS Desktop 2.12.3. Through the methodology used, it was possible to classify in the study area: 6 areas as HLZ 1, 6 areas as HLZ 2, 4 areas as HLZ 3; and 2 areas as AZ. It should be noted that an area classified as AZ covers the classifications of the other classes, and may be used as AZ, HLZ of large size (HLZ3), medium size (HLZ2) and small size helicopters (HLZ1). Likewise, an area classified as HLZ for large rotary wing aircraft (HLZ3) covers the smaller area classifications, and so on. It was concluded that images obtained through small UAV are of great use in calamity situations since they can provide data with high accuracy, with low cost, low risk and ease and agility in obtaining aerial photographs. This allows the generation, in a short time, of information about the features of the terrain in order to serve as an important decision support tool.

Keywords: disaster management, unmanned aircraft systems, helicopter landing zones, airdrop zones, random forest

Procedia PDF Downloads 146

Authors: Zuraidasahana Zulkarnain, Mohd Shafry Mohd Rahim, Nor Anita Fairos Ismail, Mohd Azhar M. Arsad

Abstract:

Handwritten signature is accepted widely as a biometric characteristic for personal authentication. The use of appropriate features plays an important role in determining accuracy of signature verification; therefore, this paper presents a feature based on the geometrical concept. To achieve the aim, triangle attributes are exploited to design a new feature since the triangle possesses orientation, angle and transformation that would improve accuracy. The proposed feature uses triangulation geometric set comprising of sides, angles and perimeter of a triangle which is derived from the center of gravity of a signature image. For classification purpose, Euclidean classifier along with Voting-based classifier is used to verify the tendency of forgery signature. This classification process is experimented using triangular geometric feature and selected global features. Based on an experiment that was validated using Grupo de Senales 960 (GPDS-960) signature database, the proposed triangular geometric feature achieves a lower Average Error Rates (AER) value with a percentage of 34% as compared to 43% of the selected global feature. As a conclusion, the proposed triangular geometric feature proves to be a more reliable feature for accurate signature verification.

Keywords: biometrics, euclidean classifier, features extraction, offline signature verification, voting-based classifier

Procedia PDF Downloads 347