Search results for: multi-class classification

Authors: Kwan Hee Han

Abstract:

In today’s business environments, frequent change of customer requirements is a tough challenge to manufacturing company. To cope with these challenges, a production planning and scheduling (PP&S) function might be established to provide accountability for both customer service and operational efficiency. Nowadays, many manufacturing firms have utilized PP&S software solutions to generate a realistic production plan and schedule to adapt to external changes efficiently. However, companies which consider the introduction of PP&S software solution, still have difficulties for selecting adequate solution to meet their specific needs. Since the task of PP&S is the one of major building blocks of SCM (Supply Chain Management) architecture, which deals with short term decision making in the production process of SCM, it is needed that the functionalities of PP&S should be analysed within the whole SCM process. The aim of this paper is to analyse the PP&S functionalities and its system architecture from the SCM perspective by using the criteria of level of planning hierarchy, major 4 SCM processes and problem-solving approaches, and finally propose a classification framework of PP&S solutions to facilitate the comparison among various commercial software solutions. By using proposed framework, several major PP&S solutions are classified and positioned according to their functional characteristics in this paper. By using this framework, practitioners who consider the introduction of computerized PP&S solutions in manufacturing firms can prepare evaluation and benchmarking sheets for selecting the most suitable solution with ease and in less time.

Keywords: production planning, production scheduling, supply chain management, the advanced planning system

Procedia PDF Downloads 182

Authors: Ilma Robo, Saimir Heta, Vera Ostreni, Elsaida Agrushi, Eduart Kapaj

Abstract:

Introduction: Cyclic hormonal fluctuations are known from literature to have a clinically visible effects on gingival tissue reactions, to the diagnosed processes of gingival inflammation. Materials and methods: At a total of 47 female patients, ad-hock presented at the University Clinic, were recorded data on effect of hormonal oscillations at periodontal treatment protocol. Oral examination was performed on soft tissue of gingiva and the oral mucous membrane, always respecting the air-drying procedure and then checking with free eye differences in oral mucosal relief. After the patients were informed about the study protocol, the purpose of the study and the ongoing procedure, verbal consensus was required. Results: The study was conducted in a total of 47 patients, out of which 13 patients were under the gingivitis classification, and 24 patients under the periodontal classification. Patients included in the study are divided by age, cycle week respectively 1,2,3 and 4.The younger age of female patients is more prone to the appearance of gingivitis, which is further aggravated by the effects of sexual hormones and the effect of the controlled or non-regulated fluctuations of the latter. Conclusions: The healing process is more fuel-intensive in the absence of high hormone levels, as they are these pro-inflammatory hormones, both in or near the ho Younger women are more open to volunteering in studies that record individual and study data that may last in time.

Keywords: gingiva, hormonal oscillations, female patients, mucosa, periodontal non-surgical treatment

Procedia PDF Downloads 64

Authors: Fernanda A. P. Peres, Thiago N. Peres, Flavio S. Fogliatto, Michel J. Anzanello

Abstract:

Batch processes are widely used in food industry and have an important role in the production of high added value products, such as chocolate. Process performance is usually described by variables that are monitored as the batch progresses. Data arising from these processes are likely to display a strong correlation-autocorrelation structure, and are usually monitored using control charts based on multiway principal components analysis (MPCA). Process control of a new batch is carried out comparing the trajectories of its relevant process variables with those in a reference set of batches that yielded products within specifications; it is clear that proper determination of the reference set is key for the success of a correct signalization of non-conforming batches in such quality control schemes. In chocolate manufacturing, misclassifications of non-conforming batches in the conching phase may lead to significant financial losses. In such context, the accuracy of process control grows in relevance. In addition to that, the main assumption in MPCA-based monitoring strategies is that all batches are synchronized in duration, both the new batch being monitored and those in the reference set. Such assumption is often not satisfied in chocolate manufacturing process. As a consequence, traditional techniques as MPCA-based charts are not suitable for process control and monitoring. To address that issue, the objective of this work is to compare the performance of three dynamic time warping (DTW) methods in the alignment and synchronization of chocolate conching process variables’ trajectories, aimed at properly determining the reference distribution for multivariate statistical process control. The power of classification of batches in two categories (conforming and non-conforming) was evaluated using the k-nearest neighbor (KNN) algorithm. Real data from a milk chocolate conching process was collected and the following variables were monitored over time: frequency of soybean lecithin dosage, rotation speed of the shovels, current of the main motor of the conche, and chocolate temperature. A set of 62 batches with durations between 495 and 1,170 minutes was considered; 53% of the batches were known to be conforming based on lab test results and experts’ evaluations. Results showed that all three DTW methods tested were able to align and synchronize the conching dataset. However, synchronized datasets obtained from these methods performed differently when inputted in the KNN classification algorithm. Kassidas, MacGregor and Taylor’s (named KMT) method was deemed the best DTW method for aligning and synchronizing a milk chocolate conching dataset, presenting 93.7% accuracy, 97.2% sensitivity and 90.3% specificity in batch classification, being considered the best option to determine the reference set for the milk chocolate dataset. Such method was recommended due to the lowest number of iterations required to achieve convergence and highest average accuracy in the testing portion using the KNN classification technique.

Keywords: batch process monitoring, chocolate conching, dynamic time warping, reference set distribution, variable duration

Procedia PDF Downloads 151

Authors: Vladimir A. Todiras

Abstract:

Spatio-temporal models provide new possibilities for real-time action in pest risk analysis. It should be noted that estimation of the possibility and probability of introduction of a pest and of its economic consequences involves many uncertainties. We present a new mapping technique that assesses pest invasion risk using online BioClass software. BioClass is a GIS tool designed to solve multiple-criteria classification and optimization problems based on fuzzy logic and level set methods. This research describes a method for predicting the potential establishment and spread of a plant pest into new areas using a case study: corn rootworm (Diabrotica spp.), tomato leaf miner (Tuta absoluta) and plum fruit moth (Grapholita funebrana). Our study demonstrated that in BioClass we can combine fuzzy logic and geographic information systems with knowledge of pest biology and environmental data to derive new information for decision making. Pests are sensitive to a warming climate, as temperature greatly affects their survival and reproductive rate and capacity. Changes have been observed in the distribution, frequency and severity of outbreaks of Helicoverpa armigera on tomato. BioClass has demonstrated to be a powerful tool for applying dynamic models and map the potential future distribution of a species, enable resource to make decisions about dangerous and invasive species management and control.

Keywords: classification, model, pest, risk

Procedia PDF Downloads 268

Authors: Paul Licina, Emma M. Johnston, David Lisle, Mark Young, Chris Brady

Abstract:

Background: Lumbar pars pathology is a common cause of pain in the growing spine. It can be seen in young athletes participating in at-risk sports and can affect sporting performance and long-term health due to its resistance to traditional management. There is a current lack of consensus of classification and treatment for pars injuries. Previous systems used CT to stage pars defects but could not assess early stress reactions. A modified classification is proposed that considers findings on MRI, significantly improving early treatment guidance. The treatment protocol is designed for patients aged 5 to 19 years. Method: Clinical screening identifies patients with a low, medium, or high index of suspicion for lumbar pars injury using patient age, sport participation and pain characteristics. MRI of the at-risk cohort enables augmentation of existing CT-based classification while avoiding ionising radiation. Patients are classified into five categories based on MRI findings. A type 0 lesion (stress reaction) is present when CT is normal and MRI shows high signal change (HSC) in the pars/pedicle on T2 images. A type 1 lesion represents the ‘early defect’ CT classification. The group previously referred to as a 'progressive stage' defect on CT can be split into 2A and 2B categories. 2As have HSC on MRI, whereas 2Bs do not. This distinction is important with regard to healing potential. Type 3 lesions are terminal stage defects on CT, characterised by pseudarthrosis. MRI shows no HSC. Results: Stress reactions (type 0) and acute fractures (1 and 2a) can heal and are treated in a custom-made hard brace for 12 weeks. It is initially worn 23 hours per day. At three weeks, patients commence basic core rehabilitation. At six weeks, in the absence of pain, the brace is removed for sleeping. Exercises are progressed to positions of daily living. Patients with continued pain remain braced 23 hours per day without exercise progression until becoming symptom-free. At nine weeks, patients commence supervised exercises out of the brace for 30 minutes each day. This allows them to re-learn muscular control without rigid support of the brace. At 12 weeks, bracing ceases and MRI is repeated. For patients with near or complete resolution of bony oedema and healing of any cortical defect, rehabilitation is focused on strength and conditioning and sport-specific exercise for the full return to activity. The length of this final stage is approximately nine weeks but depends on factors such as development and level of sports participation. If significant HSC remains on MRI, CT scan is considered to definitively assess cortical defect healing. For these patients, return to high-risk sports is delayed for up to three months. Chronic defects (2b and 3) cannot heal and are not braced, and rehabilitation follows traditional protocols. Conclusion: Appropriate clinical screening and imaging with MRI can identify pars pathology early. In those with potential for healing, we propose hard bracing and appropriate rehabilitation as part of a multidisciplinary management protocol. The validity of this protocol will be tested in future studies.

Keywords: adolescents, MRI classification, pars interticularis, treatment protocol

Procedia PDF Downloads 137

Authors: Walid Moudani

Abstract:

The healthcare environment is generally perceived as being information rich yet knowledge poor. However, there is a lack of effective analysis tools to discover hidden relationships and trends in data. In fact, valuable knowledge can be discovered from application of data mining techniques in healthcare system. In this study, a proficient methodology for the extraction of significant patterns from the coronary heart disease warehouses for heart attack prediction, which unfortunately continues to be a leading cause of mortality in the whole world, has been presented. For this purpose, we propose to enumerate dynamically the optimal subsets of the reduced features of high interest by using rough sets technique associated to dynamic programming. Therefore, we propose to validate the classification using Random Forest (RF) decision tree to identify the risky heart disease cases. This work is based on a large amount of data collected from several clinical institutions based on the medical profile of patient. Moreover, the experts’ knowledge in this field has been taken into consideration in order to define the disease, its risk factors, and to establish significant knowledge relationships among the medical factors. A computer-aided system is developed for this purpose based on a population of 525 adults. The performance of the proposed model is analyzed and evaluated based on set of benchmark techniques applied in this classification problem.

Keywords: multi-classifier decisions tree, features reduction, dynamic programming, rough sets

Procedia PDF Downloads 394

Authors: I. Stathopoulos, V. Syrgiamiotis, E. Karavasilis, A. Ploussi, I. Nikas, C. Hatzigiorgi, K. Platoni, E. P. Efstathopoulos

Abstract:

Introduction: Brain and central nervous system (CNS) tumors form the second most common group of cancer in children, accounting for 30% of all childhood cancers. MRI is the key imaging technique used for the visualization and management of pediatric brain tumors. Initial characterization of tumors from MRI scans is usually performed via a radiologist’s visual assessment. However, different brain tumor types do not always demonstrate clear differences in visual appearance. Using only conventional MRI to provide a definite diagnosis could potentially lead to inaccurate results, and so histopathological examination of biopsy samples is currently considered to be the gold standard for obtaining definite diagnoses. Machine learning is defined as the study of computational algorithms that can use, complex or not, mathematical relationships and patterns from empirical and scientific data to make reliable decisions. Concerning the above, machine learning techniques could provide effective and accurate ways to automate and speed up the analysis and diagnosis for medical images. Machine learning applications in radiology are or could potentially be useful in practice for medical image segmentation and registration, computer-aided detection and diagnosis systems for CT, MR or radiography images and functional MR (fMRI) images for brain activity analysis and neurological disease diagnosis. Purpose: The objective of this study is to provide an automated tool, which may assist in the imaging evaluation and classification of brain neoplasms in pediatric patients by determining the glioma type, grade and differentiating between different brain tissue types. Moreover, a future purpose is to present an alternative way of quick and accurate diagnosis in order to save time and resources in the daily medical workflow. Materials and Methods: A cohort, of 80 pediatric patients with a diagnosis of posterior fossa tumor, was used: 20 ependymomas, 20 astrocytomas, 20 medulloblastomas and 20 healthy children. The MR sequences used, for every single patient, were the following: axial T1-weighted (T1), axial T2-weighted (T2), FluidAttenuated Inversion Recovery (FLAIR), axial diffusion weighted images (DWI), axial contrast-enhanced T1-weighted (T1ce). From every sequence only a principal slice was used that manually traced by two expert radiologists. Image acquisition was carried out on a GE HDxt 1.5-T scanner. The images were preprocessed following a number of steps including noise reduction, bias-field correction, thresholding, coregistration of all sequences (T1, T2, T1ce, FLAIR, DWI), skull stripping, and histogram matching. A large number of features for investigation were chosen, which included age, tumor shape characteristics, image intensity characteristics and texture features. After selecting the features for achieving the highest accuracy using the least number of variables, four machine learning classification algorithms were used: k-Nearest Neighbour, Support-Vector Machines, C4.5 Decision Tree and Convolutional Neural Network. The machine learning schemes and the image analysis are implemented in the WEKA platform and MatLab platform respectively. Results-Conclusions: The results and the accuracy of images classification for each type of glioma by the four different algorithms are still on process.

Keywords: image classification, machine learning algorithms, pediatric MRI, pediatric oncology

Procedia PDF Downloads 131

Authors: Piotr Prewysz-Kwinto, Grazyna Voss

Abstract:

In June 2013, the European Parliament and the Council adopted a directive on financial reporting (Directive 2013/34/EU). The main objective was to simplify the principles of the preparation of financial statements, including the principles of the presentation and disclosures of financial information by adapting reporting burdens to the type and size of an undertaking. Therefore, the Directive introduced a classification of all undertakings into five groups, i.e. micro, small, medium-sized, large and public-interest entities, and defined in detail the classification criteria. The principles of the preparation of financial statements and the presentation of financial information as well as applicable simplifications were defined for each group. The EU Member States had to implement the provisions of Directive 34 relating to accounting and financial reporting into domestic norms until January 1, 2016. In Poland, the provisions of Directive 34 were implemented into domestic accounting norms specified in the Polish Accounting Act on a gradual basis. On July 11, 2014, the Polish Parliament adopted an amendment to the Act, introducing the Directive's solutions for micro-undertakings and on July 23, 2015, for the remaining undertakings. The aim of this paper is to present Polish solutions relating to financial reporting after the implementation of Directive 34 and the results of the survey conducted among accountants regarding the evaluation of the implemented simplifications for micro and small undertakings.

Keywords: accounting standards, financial reporting, financial statement, simplification

Procedia PDF Downloads 264

Authors: Ahmad Shahin, Walid Moudani, Ali Bekraki

Abstract:

This research aims at presenting a predictive data mining model to handle an accurate diagnosis of acute appendicitis with patients for the purpose of maximizing the health service quality, minimizing morbidity/mortality, and reducing cost. However, acute appendicitis is the most common disease which requires timely accurate diagnosis and needs surgical intervention. Although the treatment of acute appendicitis is simple and straightforward, its diagnosis is still difficult because no single sign, symptom, laboratory or image examination accurately confirms the diagnosis of acute appendicitis in all cases. This contributes in increasing morbidity and negative appendectomy. In this study, the authors propose to generate an accurate model in prediction of patients with acute appendicitis which is based, firstly, on the segmentation technique associated to ABC algorithm to segment the patients; secondly, on applying fuzzy logic to process the massive volume of heterogeneous and noisy data (age, sex, fever, white blood cell, neutrophilia, CRP, urine, ultrasound, CT, appendectomy, etc.) in order to express knowledge and analyze the relationships among data in a comprehensive manner; and thirdly, on applying dynamic programming technique to reduce the number of data attributes. The proposed model is evaluated based on a set of benchmark techniques and even on a set of benchmark classification problems of osteoporosis, diabetes and heart obtained from the UCI data and other data sources.

Keywords: healthcare management, acute appendicitis, data mining, classification, decision tree

Procedia PDF Downloads 333

Authors: Jonathan Sohn

Abstract:

Deep neural networks (DNNs) have shown state-of-the-art performance for many applications, including computer vision, natural language processing, and speech recognition. Recently, adversarial attacks have been studied in the context of deep neural networks, which aim to alter the results of deep neural networks by modifying the inputs slightly. For example, an adversarial attack on a DNN used for object detection can cause the DNN to miss certain objects. As a result, the reliability of DNNs is undermined by their lack of robustness against adversarial attacks, raising concerns about their use in safety-critical applications such as autonomous driving. In this paper, we focus on studying the adversarial attacks and defenses on DNNs for image classification. There are two types of adversarial attacks studied which are fast gradient sign method (FGSM) attack and projected gradient descent (PGD) attack. A DNN forms decision boundaries that separate the input images into different categories. The adversarial attack slightly alters the image to move over the decision boundary, causing the DNN to misclassify the image. FGSM attack obtains the gradient with respect to the image and updates the image once based on the gradients to cross the decision boundary. PGD attack, instead of taking one big step, repeatedly modifies the input image with multiple small steps. There is also another type of attack called the target attack. This adversarial attack is designed to make the machine classify an image to a class chosen by the attacker. We can defend against adversarial attacks by incorporating adversarial examples in training. Specifically, instead of training the neural network with clean examples, we can explicitly let the neural network learn from the adversarial examples. In our experiments, the digit recognition accuracy on the MNIST dataset drops from 97.81% to 39.50% and 34.01% when the DNN is attacked by FGSM and PGD attacks, respectively. If we utilize FGSM training as a defense method, the classification accuracy greatly improves from 39.50% to 92.31% for FGSM attacks and from 34.01% to 75.63% for PGD attacks. To further improve the classification accuracy under adversarial attacks, we can also use a stronger PGD training method. PGD training improves the accuracy by 2.7% under FGSM attacks and 18.4% under PGD attacks over FGSM training. It is worth mentioning that both FGSM and PGD training do not affect the accuracy of clean images. In summary, we find that PGD attacks can greatly degrade the performance of DNNs, and PGD training is a very effective way to defend against such attacks. PGD attacks and defence are overall significantly more effective than FGSM methods.

Keywords: deep neural network, adversarial attack, adversarial defense, adversarial machine learning

Procedia PDF Downloads 175

Authors: Mohamed Al-Badrashiny, Abdelghani Bellaachia

Abstract:

This paper presents a classifier ensemble approach for predicting the survivability of the breast cancer patients using the latest database version of the Surveillance, Epidemiology, and End Results (SEER) Program of the National Cancer Institute. The system consists of two main components; features selection and classifier ensemble components. The features selection component divides the features in SEER database into four groups. After that it tries to find the most important features among the four groups that maximizes the weighted average F-score of a certain classification algorithm. The ensemble component uses three different classifiers, each of which models different set of features from SEER through the features selection module. On top of them, another classifier is used to give the final decision based on the output decisions and confidence scores from each of the underlying classifiers. Different classification algorithms have been examined; the best setup found is by using the decision tree, Bayesian network, and Na¨ıve Bayes algorithms for the underlying classifiers and Na¨ıve Bayes for the classifier ensemble step. The system outperforms all published systems to date when evaluated against the exact same data of SEER (period of 1973-2002). It gives 87.39% weighted average F-score compared to 85.82% and 81.34% of the other published systems. By increasing the data size to cover the whole database (period of 1973-2014), the overall weighted average F-score jumps to 92.4% on the held out unseen test set.

Keywords: classifier ensemble, breast cancer survivability, data mining, SEER

Procedia PDF Downloads 304

Authors: Esra Polat

Abstract:

Partial Least Squares Discriminant Analysis (PLSDA) is a statistical method for classification and consists a classical Partial Least Squares Regression (PLSR) in which the dependent variable is a categorical one expressing the class membership of each observation. PLSDA can be applied in many cases when classical discriminant analysis cannot be applied. For example, when the number of observations is low and when the number of independent variables is high. When there are missing values, PLSDA can be applied on the data that is available. Finally, it is adapted when multicollinearity between independent variables is high. The aim of this study is to determine the economic and/or demographic indicators, which are effective in grouping the 28 European Union (EU) member countries and 7 candidate countries (including potential candidates Bosnia and Herzegovina (BiH) and Kosova) by using the data set obtained from database of the World Bank for 2014. Leaving the political issues aside, the analysis is only concerned with the economic and demographic variables that have the potential influence on country’s eligibility for EU entrance. Hence, in this study, both the performance of PLSDA method in classifying the countries correctly to their pre-defined groups (candidate or member) and the differences between the EU countries and candidate countries in terms of these indicators are analyzed. As a result of the PLSDA, the value of percentage correctness of 100 % indicates that overall of the 35 countries is classified correctly. Moreover, the most important variables that determine the statuses of member and candidate countries in terms of economic indicators are identified as 'external balance on goods and services (% GDP)', 'gross domestic savings (% GDP)' and 'gross national expenditure (% GDP)' that means for the 2014 economical structure of countries is the most important determinant of EU membership. Subsequently, the model validated to prove the predictive ability by using the data set for 2015. For prediction sample, %97,14 of the countries are correctly classified. An interesting result is obtained for only BiH, which is still a potential candidate for EU, predicted as a member of EU by using the indicators data set for 2015 as a prediction sample. Although BiH has made a significant transformation from a war-torn country to a semi-functional state, ethnic tensions, nationalistic rhetoric and political disagreements are still evident, which inhibit Bosnian progress towards the EU.

Keywords: classification, demographic indicators, economic indicators, European Union, partial least squares discriminant analysis

Procedia PDF Downloads 265

Authors: Anoop T. R., Otman Basir, Robert F. Hess, Eileen E. Birch, Brooke A. Koritala, Reed M. Jost, Becky Luu, David Stager, Ben Thompson

Abstract:

Strabismus refers to a misalignment of the eyes. Early detection and treatment of strabismus in childhood can prevent the development of permanent vision loss due to abnormal development of visual brain areas. We developed a two-stage method for strabismus detection and classification based on photographs of the face. The first stage detects the presence or absence of strabismus, and the second stage classifies the type of strabismus. The first stage comprises face detection using Haar cascade, facial landmark estimation, face alignment, aligned face landmark detection, segmentation of the eye region, and detection of strabismus using VGG 16 convolution neural networks. Face alignment transforms the face to a canonical pose to ensure consistency in subsequent analysis. Using facial landmarks, the eye region is segmented from the aligned face and fed into a VGG 16 CNN model, which has been trained to classify strabismus. The CNN determines whether strabismus is present and classifies the type of strabismus (exotropia, esotropia, and vertical deviation). If stage 1 detects strabismus, the eye region image is fed into stage 2, which starts with the estimation of pupil center coordinates using mask R-CNN deep neural networks. Then, the distance between the pupil coordinates and eye landmarks is calculated along with the angle that the pupil coordinates make with the horizontal and vertical axis. The distance and angle information is used to characterize the degree and direction of the strabismic eye misalignment. This model was tested on 100 clinically labeled images of children with (n = 50) and without (n = 50) strabismus. The True Positive Rate (TPR) and False Positive Rate (FPR) of the first stage were 94% and 6% respectively. The classification stage has produced a TPR of 94.73%, 94.44%, and 100% for esotropia, exotropia, and vertical deviations, respectively. This method also had an FPR of 5.26%, 5.55%, and 0% for esotropia, exotropia, and vertical deviation, respectively. The addition of one more feature related to the location of corneal light reflections may reduce the FPR, which was primarily due to children with pseudo-strabismus (the appearance of strabismus due to a wide nasal bridge or skin folds on the nasal side of the eyes).

Keywords: strabismus, deep neural networks, face detection, facial landmarks, face alignment, segmentation, VGG 16, mask R-CNN, pupil coordinates, angle deviation, horizontal and vertical deviation

Procedia PDF Downloads 63

Authors: Rongbo Shen, Jianhua Yao, Kezhou Yan, Kuan Tian, Cheng Jiang, Ke Zhou

Abstract:

This study proposes a generative adversarial networks (GAN) framework to perform synthetic sampling in feature space, i.e., feature augmentation, to address the class imbalance problem in medical image analysis. A feature extraction network is first trained to convert images into feature space. Then the GAN framework incorporates adversarial learning to train a feature generator for the minority class through playing a minimax game with a discriminator. The feature generator then generates features for minority class from arbitrary latent distributions to balance the data between the majority class and the minority class. Additionally, a data cleaning technique, i.e., Tomek link, is employed to clean up undesirable conflicting features introduced from the feature augmentation and thus establish well-defined class clusters for the training. The experiment section evaluates the proposed method on two medical image analysis tasks, i.e., mass classification on mammogram and cancer metastasis classification on histopathological images. Experimental results suggest that the proposed method obtains superior or comparable performance over the state-of-the-art counterparts. Compared to all counterparts, our proposed method improves more than 1.5 percentage of accuracy.

Keywords: class imbalance, synthetic sampling, feature augmentation, generative adversarial networks, data cleaning

Procedia PDF Downloads 111

Authors: Magdalena Igras, Joanna Grzybowska, Mariusz Ziółko

Abstract:

The study of emotions expressed in emergency phone call is presented, covering both statistical analysis of emotions configurations and an attempt to automatically classify emotions. An emergency call is a situation usually accompanied by intense, authentic emotions. They influence (and may inhibit) the communication between caller and responder. In order to support responders in their responsible and psychically exhaustive work, we studied when and in which combinations emotions appeared in calls. A corpus of 45 hours of conversations (about 3300 calls) from emergency call center was collected. Each recording was manually tagged with labels of emotions valence (positive, negative or neutral), type (sadness, tiredness, anxiety, surprise, stress, anger, fury, calm, relief, compassion, satisfaction, amusement, joy) and arousal (weak, typical, varying, high) on the basis of perceptual judgment of two annotators. As we concluded, basic emotions tend to appear in specific configurations depending on the overall situational context and attitude of speaker. After performing statistical analysis we distinguished four main types of emotional behavior of callers: worry/helplessness (sadness, tiredness, compassion), alarm (anxiety, intense stress), mistake or neutral request for information (calm, surprise, sometimes with amusement) and pretension/insisting (anger, fury). The frequency of profiles was respectively: 51%, 21%, 18% and 8% of recordings. A model of presenting the complex emotional profiles on the two-dimensional (tension-insecurity) plane was introduced. In the stage of acoustic analysis, a set of prosodic parameters, as well as Mel-Frequency Cepstral Coefficients (MFCC) were used. Using these parameters, complex emotional states were modeled with machine learning techniques including Gaussian mixture models, decision trees and discriminant analysis. Results of classification with several methods will be presented and compared with the state of the art results obtained for classification of basic emotions. Future work will include optimization of the algorithm to perform in real time in order to track changes of emotions during a conversation.

Keywords: acoustic analysis, complex emotions, emotion recognition, machine learning

Procedia PDF Downloads 377

Authors: M. Mueller, M. Kuehn, M. Voelker

Abstract:

In small and medium-sized enterprises (SMEs), the challenge is to create a well-grounded and reliable basis for process analysis, optimization and planning due to a lack of data. SMEs have limited access to methods with which they can effectively and efficiently analyse processes and identify cause-and-effect relationships in order to generate the necessary database and derive optimization potential from it. The implementation of digitalization within the framework of Industry 4.0 thus becomes a particular necessity for SMEs. For these reasons, the abstract presents an analysis methodology that is subject to the objective of developing an SME-appropriate methodology for efficient, temporarily feasible data collection and evaluation in flexible production and logistics systems as a basis for process analysis and optimization. The overall methodology focuses on retrospective, event-based tracing and analysis of material flow objects. The technological basis consists of Bluetooth low energy (BLE)-based transmitters, so-called beacons, and smart mobile devices (SMD), e.g. smartphones as receivers, between which distance data can be measured and derived motion profiles. The distance is determined using the Received Signal Strength Indicator (RSSI), which is a measure of signal field strength between transmitter and receiver. The focus is the development of a software-based methodology for interpretation of relative movements of transmitters and receivers based on distance data. The main research is on selection and implementation of pattern recognition methods for automatic process recognition as well as methods for the visualization of relative distance data. Due to an existing categorization of the database regarding process types, classification methods (e.g. Support Vector Machine) from the field of supervised learning are used. The necessary data quality requires selection of suitable methods as well as filters for smoothing occurring signal variations of the RSSI, the integration of methods for determination of correction factors depending on possible signal interference sources (columns, pallets) as well as the configuration of the used technology. The parameter settings on which respective algorithms are based have a further significant influence on result quality of the classification methods, correction models and methods for visualizing the position profiles used. The accuracy of classification algorithms can be improved up to 30% by selected parameter variation; this has already been proven in studies. Similar potentials can be observed with parameter variation of methods and filters for signal smoothing. Thus, there is increased interest in obtaining detailed results on the influence of parameter and factor combinations on data quality in this area. The overall methodology is realized with a modular software architecture consisting of independently modules for data acquisition, data preparation and data storage. The demonstrator for initialization and data acquisition is available as mobile Java-based application. The data preparation, including methods for signal smoothing, are Python-based with the possibility to vary parameter settings and to store them in the database (SQLite). The evaluation is divided into two separate software modules with database connection: the achievement of an automated assignment of defined process classes to distance data using selected classification algorithms and the visualization as well as reporting in terms of a graphical user interface (GUI).

Keywords: event-based tracing, machine learning, process classification, parameter settings, RSSI, signal smoothing

Procedia PDF Downloads 111

Authors: Harsha B. R., K. S. Anil Kumar

Abstract:

Land characterization, classification, and soil suitability evaluation of grapes-growing pedons were assessed at fifteen taluks covering four agro climatic zones of Karnataka. Study on problems and potentials of grapes cultivation in selected agro-climatic zones was carried out along with the plant sample analysis. Twenty soil profiles were excavated as study site based on the dominance of area falling under grapes production and existing spatial variability of soils. The detailed information of profiles and horizon wise soil samples were collected to study the morphological, physical, chemical, and fertility characteristics. Climatic analysis and water retention characteristics of soils of major grapes-growing areas were also done. Based on the characterisation and classification study, it was revealed that soils of Doddaballapur (Bangalore Blue and Wine grapes), Bangalore North (GKVK Farm, Rajankunte, and IIHR Farm), Devanahalli, Magadi, Hoskote, Chikkaballapur (Dilkush and Red globe), Yelaburga, Hagari Bommanahalli, Bagalkot (UHS farm) and Indi fall under the soil order Alfisol. Vijaypur pedon of northern dry zone was keyed out as Vertisols whereas, Jamkhandi and Athani as Inceptisols. Properties of Aridisols were observed in B. Bagewadi (Manikchaman and Thompson Seedless) and Afzalpur. Soil fertility status and its mapping using GIS technique revealed that all the nutrients were found to be in adequate range except nitrogen, potassium, zinc, iron, and boron, which indicated the need for application along with organic matter to improve the SOC status. Varieties differed among themselves in yield and plant nutrient composition depending on their age, climatic, soil, and management requirements. Bangalore North (GKVK farm) and Jamkhandi are having medium soil organic carbon stocks of 6.21 and 6.55 kg m⁻³, respectively. Soils of Bangalore North (Rajankunte) were highly suitable (S1) for grapes cultivation. Under northern Karnataka, Vijayapura, B. Bagewadi, Indi, and Afzalpur vineyards were good performers despite the limitations of fertility and free lime content.

Keywords: land characterization, suitability, soil orders, soil organic carbon stock

Procedia PDF Downloads 92

Authors: Nassima Noufail, Sara Bouhali

Abstract:

In this work, we develop a semi-supervised solution for the purpose of action detection in videos and propose an efficient algorithm for video segmentation. The approach is divided into video segmentation, feature extraction, and classification. In the first part, a video is segmented into clips, and we used the K-means algorithm for this segmentation; our goal is to find groups based on similarity in the video. The application of k-means clustering into all the frames is time-consuming; therefore, we started by the identification of transition frames where the scene in the video changes significantly, and then we applied K-means clustering into these transition frames. We used two image filters, the gaussian filter and the Laplacian of Gaussian. Each filter extracts a set of features from the frames. The Gaussian filter blurs the image and omits the higher frequencies, and the Laplacian of gaussian detects regions of rapid intensity changes; we then used this vector of filter responses as an input to our k-means algorithm. The output is a set of cluster centers. Each video frame pixel is then mapped to the nearest cluster center and painted with a corresponding color to form a visual map. The resulting visual map had similar pixels grouped. We then computed a cluster score indicating how clusters are near each other and plotted a signal representing frame number vs. clustering score. Our hypothesis was that the evolution of the signal would not change if semantically related events were happening in the scene. We marked the breakpoints at which the root mean square level of the signal changes significantly, and each breakpoint is an indication of the beginning of a new video segment. In the second part, for each segment from part 1, we randomly selected a 16-frame clip, then we extracted spatiotemporal features using convolutional 3D network C3D for every 16 frames using a pre-trained model. The C3D final output is a 512-feature vector dimension; hence we used principal component analysis (PCA) for dimensionality reduction. The final part is the classification. The C3D feature vectors are used as input to a multi-class linear support vector machine (SVM) for the training model, and we used a multi-classifier to detect the action. We evaluated our experiment on the UCF101 dataset, which consists of 101 human action categories, and we achieved an accuracy that outperforms the state of art by 1.2%.

Keywords: video segmentation, action detection, classification, Kmeans, C3D

Procedia PDF Downloads 58

Authors: K. Petcharaporn

Abstract:

The external characteristics of tomatoes, such as freshness, color and size are typically used in quality control processes for tomatoes sorting. However, the internal mold infection of intact tomato cannot be sorted based on a visible assessment and destructive method alone. In this study, a non-destructive technique was used to predict the internal mold infection of intact tomatoes by using transmittance visible and near infrared (VIS-NIR) spectroscopy. Spectra for 200 samples contained 100 samples for normal tomatoes and 100 samples for mold infected tomatoes were acquired in the wavelength range between 665-955 nm. This data was used in conjunction with partial least squares-discriminant analysis (PLS-DA) method to generate a classification model for tomato quality between groups of internal mold infection of intact tomato samples. For this task, the data was split into two groups, 140 samples were used for a training set and 60 samples were used for a test set. The spectra of both normal and internally mold infected tomatoes showed different features in the visible wavelength range. Combined spectral pretreatments of standard normal variate transformation (SNV) and smoothing (Savitzky-Golay) gave the optimal calibration model in training set, 85.0% (63 out of 71 for the normal samples and 56 out of 69 for the internal mold samples). The classification accuracy of the best model on the test set was 91.7% (29 out of 29 for the normal samples and 26 out of 31 for the internal mold tomato samples). The results from this experiment showed that transmittance VIS-NIR spectroscopy can be used as a non-destructive technique to predict the internal mold infection of intact tomatoes.

Keywords: tomato, mold, quality, prediction, transmittance

Procedia PDF Downloads 350

Authors: Atefeh Heydari, Mohammadali Tavakoli, Naomie Salim

Abstract:

In recent years, we have witnessed that online reviews are the most important source of customers’ opinion. They are progressively more used by individuals and organisations to make purchase and business decisions. Unfortunately, for the reason of profit or fame, frauds produce deceptive reviews to hoodwink potential customers. Their activities mislead not only potential customers to make appropriate purchasing decisions and organisations to reshape their business, but also opinion mining techniques by preventing them from reaching accurate results. Spam reviews could be divided into two main groups, i.e. multiple and singleton spam reviews. Detecting a singleton spam review that is the only review written by a user ID is extremely challenging due to lack of clue for detection purposes. Singleton spam reviews are very harmful and various features and proofs used in multiple spam reviews detection are not applicable in this case. Current research aims to propose a novel supervised technique to detect singleton spam reviews. To achieve this, various features are proposed in this study and are to be combined with the most appropriate features extracted from literature and employed in a classifier. In order to compare the performance of different classifiers, SVM and naive Bayes classification algorithms were used for model building. The results revealed that SVM was more accurate than naive Bayes and our proposed technique is capable to detect singleton spam reviews effectively.

Keywords: classification algorithms, Naïve Bayes, opinion review spam detection, singleton review spam detection, support vector machine

Procedia PDF Downloads 286

Authors: David Percy

Abstract:

Landslides are a geologic phenomenon that affects a large number of inhabited places and are constantly being monitored and studied for the prediction of future occurrences. Reconstructability analysis (RA) is a methodology for extracting informative models from large volumes of data that work exclusively with discrete data. While RA has been used in medical applications and social science extensively, we are introducing it to the spatial sciences through applications like landslide prediction. Since RA works exclusively with discrete data, such as soil classification or bedrock type, working with continuous data, such as porosity, requires that these data are binned for inclusion in the model. RA constructs models of the data which pick out the most informative elements, independent variables (IVs), from each layer that predict the dependent variable (DV), landslide occurrence. Each layer included in the model retains its classification data as a primary encoding of the data. Unlike other machine learning algorithms that force the data into one-hot encoding type of schemes, RA works directly with the data as it is encoded, with the exception of continuous data, which must be binned. The usual physical and derived layers are included in the model, and testing our results against other published methodologies, such as neural networks, yields accuracy that is similar but with the advantage of a completely transparent model. The results of an RA session with a data set are a report on every combination of variables and their probability of landslide events occurring. In this way, every combination of informative state combinations can be examined.

Keywords: reconstructability analysis, machine learning, landslides, raster analysis

Procedia PDF Downloads 46

Authors: K. Petcharaporn, N. Prathengjit

Abstract:

The external characteristics of tomatoes, such as freshness, color and size are typically used in quality control processes for tomatoes sorting. However, the internal mold infection of intact tomato cannot be sorted based on a visible assessment and destructive method alone. In this study, a non-destructive technique was used to predict the internal mold infection of intact tomatoes by using transmittance visible and near infrared (VIS-NIR) spectroscopy. Spectra for 200 samples contained 100 samples for normal tomatoes and 100 samples for mold infected tomatoes were acquired in the wavelength range between 665-955 nm. This data was used in conjunction with partial least squares-discriminant analysis (PLS-DA) method to generate a classification model for tomato quality between groups of internal mold infection of intact tomato samples. For this task, the data was split into two groups, 140 samples were used for a training set and 60 samples were used for a test set. The spectra of both normal and internally mold infected tomatoes showed different features in the visible wavelength range. Combined spectral pretreatments of standard normal variate transformation (SNV) and smoothing (Savitzky-Golay) gave the optimal calibration model in training set, 85.0% (63 out of 71 for the normal samples and 56 out of 69 for the internal mold samples). The classification accuracy of the best model on the test set was 91.7% (29 out of 29 for the normal samples and 26 out of 31 for the internal mold tomato samples). The results from this experiment showed that transmittance VIS-NIR spectroscopy can be used as a non-destructive technique to predict the internal mold infection of intact tomatoes.

Keywords: tomato, mold, quality, prediction, transmittance

Procedia PDF Downloads 500

Authors: T. Garba, Y. Y. Sabo A. Babanyara, K. G. Ilellah, A. K. Mutari

Abstract:

This study used the Normalized Difference Vegetation Index (NDVI) and maps compiled from the classification of Landsat TM and Landsat ETM images of 1986 and 1999 respectively and Nigeria sat 1 images of 2007 to quantify changes in land use and land cover in selected areas of Nigeria covering 143,609 hectares that are threatened by the encroaching Sahara desert. The results of this investigation revealed a decrease in natural vegetation over the three time slices (1986, 1999 and 2007) which was characterised by an increase in high positive pixel values from 0.04 in 1986 to 0.22 and 0.32 in 1999 and 2007 respectively and, a decrease in natural vegetation from 74,411.60ha in 1986 to 28,591.93ha and 21,819.19ha in 1999 and 2007 respectively. The same results also revealed a periodic trend in which there was progressive increase in the cultivated area from 60,191.87ha in 1986 to 104,376.07ha in 1999 and a terminal decrease to 88,868.31ha in 2007. These findings point to expansion of vegetated and cultivated areas in in the initial period between 1988 and 1996 and reversal of these increases in the terminal period between 1988 and 1996. The study also revealed progressive expansion of built-up areas from 1, 681.68ha in 1986 to 2,661.82ha in 1999 and to 3,765.35ha in 2007. These results argue for the urgent need to protect and conserve the depleting natural vegetation by adopting sustainable human resource use practices i.e. intensive farming in order to minimize persistent depletion of natural vegetation.

Keywords: changes, classification, desertification, vegetation changes

Procedia PDF Downloads 371

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

Abstract:

In this study, it was tried to identify some heart rhythm disorders by electrocardiography (ECG) data that is taken from MIT-BIH arrhythmia database by subtracting the required features, presenting to artificial neural networks (ANN), artificial immune systems (AIS), artificial neural network based on artificial immune system (AIS-ANN) and particle swarm optimization based artificial neural network (PSO-NN) classifier systems. The main purpose of this study is to evaluate the performance of hybrid AIS-ANN and PSO-ANN classifiers with regard to the ANN and AIS. For this purpose, the normal sinus rhythm (NSR), atrial premature contraction (APC), sinus arrhythmia (SA), ventricular trigeminy (VTI), ventricular tachycardia (VTK) and atrial fibrillation (AF) data for each of the RR intervals were found. Then these data in the form of pairs (NSR-APC, NSR-SA, NSR-VTI, NSR-VTK and NSR-AF) is created by combining discrete wavelet transform which is applied to each of these two groups of data and two different data sets with 9 and 27 features were obtained from each of them after data reduction. Afterwards, the data randomly was firstly mixed within themselves, and then 4-fold cross validation method was applied to create the training and testing data. The training and testing accuracy rates and training time are compared with each other. As a result, performances of the hybrid classification systems, AIS-ANN and PSO-ANN were seen to be close to the performance of the ANN system. Also, the results of the hybrid systems were much better than AIS, too. However, ANN had much shorter period of training time than other systems. In terms of training times, ANN was followed by PSO-ANN, AIS-ANN and AIS systems respectively. Also, the features that extracted from the data affected the classification results significantly.

Keywords: AIS, ANN, ECG, hybrid classifiers, PSO

Procedia PDF Downloads 423

Authors: Yeliz Karaca, Carlo Cattani

Abstract:

This study focuses on the local Holder exponent as a measure of the function regularity for time series related to finance data. In this study, the attributes of the finance dataset belonging to 13 countries (India, China, Japan, Sweden, France, Germany, Italy, Australia, Mexico, United Kingdom, Argentina, Brazil, USA) located in 5 different continents (Asia, Europe, Australia, North America and South America) have been examined.These countries are the ones mostly affected by the attributes with regard to financial development, covering a period from 2012 to 2017. Our study is concerned with the most important attributes that have impact on the development of finance for the countries identified. Our method is comprised of the following stages: (a) among the multi fractal methods and Brownian motion Holder regularity functions (polynomial, exponential), significant and self-similar attributes have been identified (b) The significant and self-similar attributes have been applied to the Artificial Neuronal Network (ANN) algorithms (Feed Forward Back Propagation (FFBP) and Cascade Forward Back Propagation (CFBP)) (c) the outcomes of classification accuracy have been compared concerning the attributes that have impact on the attributes which affect the countries’ financial development. This study has enabled to reveal, through the application of ANN algorithms, how the most significant attributes are identified within the relevant dataset via the Holder functions (polynomial and exponential function).

Keywords: artificial neural networks, finance data, Holder regularity, multifractals

Procedia PDF Downloads 230

Authors: Sushma Ghogale

Abstract:

With a surge in user-generated content or feedback or reviews on the internet, it has become possible and important to know consumers' opinions about products and services. This data is important for both potential customers and businesses providing the services. Data from social media is attracting significant attention and has become the most prominent channel of expressing an unregulated opinion. Prospective customers look for reviews from experienced customers before deciding to buy a product or service. Several websites provide a platform for users to post their feedback for the provider and potential customers. However, the biggest challenge in analyzing such data is in extracting latent features and providing term-level analysis of the data. This paper proposes an approach to use topic modeling to classify the reviews into topics and conduct sentiment analysis to mine the opinions. This approach can analyse and classify latent topics mentioned by reviewers on business sites or review sites, or social media using topic modeling to identify the importance of each topic. It is followed by sentiment analysis to assess the satisfaction level of each topic. This approach provides a classification of hotel reviews using multiple machine learning techniques and comparing different classifiers to mine the opinions of user reviews through sentiment analysis. This experiment concludes that Multinomial Naïve Bayes classifier produces higher accuracy than other classifiers.

Keywords: latent Dirichlet allocation, topic modeling, text classification, sentiment analysis

Procedia PDF Downloads 86

Authors: Zegrar Ahmed, Ghabi Mohamed

Abstract:

The degradation of arid and semi-arid ecosystems in Algeria has become a palpable fact that only hinders progress and rural development. In these exceptionally fragile environments, the decline of vegetation is done according to an alarming increase and wind erosion dominates. The ecosystem is subjected to a long hot dry season and low annual average rainfall. The urgency of the fight against desertification is imposed by the very nature of the process that tends to self-accelerate, resulting when human intervention is not forthcoming the irreversibility situations, preventing any possibility of restoration state of these zones. These phenomena have led to different degradation processes, such as the destruction of vegetation, soil erosion, and deterioration of the physical environment. In this study, the work is mainly based on the criteria for classification and identification of physical parameters for spatial analysis and multi-sources to determine the vulnerability of major steppe formations and their impact on desertification. we used Landsat data with two different dates March 2010 and November 2014 in order to determine the changes in land cover, sand moving and land degradation for the diagnosis of the desertification Phenomenon. The application, through specific processes, including the supervised classification was used to characterize the main steppe formations. An analysis of the vulnerability of plant communities was conducted to assign weights and identify areas most susceptible to desertification. Vegetation indices are used to characterize the steppe formations to determine changes in land use.

Keywords: remote sensing, SIG, ecosystem, degradation, desertification

Procedia PDF Downloads 325

Authors: Kwaku Damoah

Abstract:

This paper presents a methodology and software application (App) designed to empower users in accessing, retrieving, and comparatively exploring data within the hierarchical network framework of the Government Finance Statistics (GFS) system. It explores the ease of navigating the GFS system and identifies the gaps filled by the new methodology and App. The GFS, embodies a complex Hierarchical Network Classification (HNC) structure, encapsulating institutional units, revenues, expenses, assets, liabilities, and economic activities. Navigating this structure demands specialized knowledge, experience, and skill, posing a significant challenge for effective analytics and fiscal policy decision-making. Many professionals encounter difficulties deciphering these classifications, hindering confident utilization of the system. This accessibility barrier obstructs a vast number of professionals, students, policymakers, and the public from leveraging the abundant data and information within the GFS. Leveraging R programming language, Data Science Analytics and Machine Learning, an efficient methodology enabling users to access, navigate, and conduct exploratory comparisons was developed. The machine learning Fiscal Analytics App (FLOWZZ) democratizes access to advanced analytics through its user-friendly interface, breaking down expertise barriers.

Keywords: data science, data wrangling, drilldown analytics, government finance statistics, hierarchical network classification, machine learning, web application.

Procedia PDF Downloads 48

Authors: Anitha Muralidhara, Victor Engelen, Christophe Len, Pascal Pandard, Guy Marlair

Abstract:

Production of bio-based chemicals and materials from lignocellulosic biomass is gaining tremendous importance in advanced bio-refineries while aiming towards progressive replacement of petroleum based chemicals in transportation fuels and commodity polymers. One such attempt has resulted in the production of key furan derivatives (FD) such as furfural, HMF, MMF etc., via acid catalyzed dehydration (ACD) of C6 and C5 sugars, which are further converted into key chemicals or intermediates (such as Furandicarboxylic acid, Furfuryl alcohol etc.,). In subsequent processes, many high potential FD are produced, that can be converted into high added value polymers or high energy density biofuels. During ACD, an unavoidable polyfuranic byproduct is generated which is called humins. The family of FD is very large with varying chemical structures and diverse physicochemical properties. Accordingly, the associated risk profiles may largely vary. Hazardous Material (Haz-mat) classification systems such as GHS (CLP in the EU) and the UN TDG Model Regulations for transport of dangerous goods are one of the preliminary requirements for all chemicals for their appropriate classification, labelling, packaging, safe storage, and transportation. Considering the growing application routes of FD, it becomes important to notice the limited access to safety related information (safety data sheets available only for famous compounds such as HMF, furfural etc.,) in these internationally recognized haz-mat classification systems. However, these classifications do not necessarily provide information about the extent of risk involved when the chemical is used in any specific application. Factors such as thermal stability, speed of combustion, chemical incompatibilities, etc., can equally influence the safety profile of a compound, that are clearly out of the scope of any haz-mat classification system. Irrespective of the bio-based origin, FD has so far received inconsistent remarks concerning their toxicity profiles. With such inconsistencies, there is a fear that, a large family of FD may also follow extreme judgmental scenarios like ionic liquids, by ranking some compounds as extremely thermally stable, non-flammable, etc., Unless clarified, these messages could lead to misleading judgements while ranking the chemical based on its hazard rating. Safety is a key aspect in any sustainable biorefinery operation/facility, which is often underscored or neglected. To fill up these existing data gaps and to address ambiguities and discrepancies, the current study focuses on giving preliminary insights on safety assessment of FD and their potential targeted by-products. With the available information in the literature and obtained experimental results, physicochemical safety, environmental safety as well as (a scenario based) fire safety profiles of key FD, as well as side streams such as humins and levulinic acid, will be considered. With this, the study focuses on defining patterns and trends that gives coherent safety related information for existing and newly synthesized FD in the market for better functionality and sustainable applications.

Keywords: furanics, humins, safety, thermal and fire hazard, toxicity

Procedia PDF Downloads 153

Authors: Dmitriy Babenko, Sergey Yegorov, Ilya Korshukov, Aidana Sultanbekova, Valentina Barkhanskaya, Tatiana Bashirova, Yerzhan Zhunusov, Yevgeniya Li, Viktoriya Parakhina, Svetlana Kolesnichenko, Yeldar Baiken, Aruzhan Pralieva, Zhibek Zhumadilova, Matthew S. Miller, Gonzalo H. Hortelano, Anar Turmuhambetova, Antonella E. Chesca, Irina Kadyrova

Abstract:

Introduction: The rapidly evolving COVID-19 pandemic, along with the re-emergence of pathogens causing acute respiratory infections (ARI), has necessitated the development of novel diagnostic tools to differentiate various causes of ARI. MALDI-MS, due to its wide usage and affordability, has been proposed as a potential instrument for diagnosing SARS-CoV-2 versus non-SARS-CoV-2 ARI. The aim of this study was to investigate the potential of MALDI-MS in conjunction with a machine learning model to accurately distinguish between symptomatic infections caused by SARS-CoV-2 and non-SARS-CoV-2 during both the early and later phases of the pandemic. Furthermore, this study aimed to analyze mass spectrometry (MS) data obtained from nasal swabs of healthy individuals. Methods: We gathered mass spectra from 252 samples, comprising 108 SARS-CoV-2-positive samples obtained in 2020 (Covid 2020), 7 SARS-CoV- 2-positive samples obtained in 2023 (Covid 2023), 71 samples from symptomatic individuals without SARS-CoV-2 (Control non-Covid ARVI), and 66 samples from healthy individuals (Control healthy). All the samples were subjected to RT-PCR testing. For data analysis, we employed the caret R package to train and test seven machine-learning algorithms: C5.0, KNN, NB, RF, SVM-L, SVM-R, and XGBoost. We conducted a training process using a five-fold (outer) nested repeated (five times) ten-fold (inner) cross-validation with a randomized stratified splitting approach. Results: In this study, we utilized the Covid 2020 dataset as a case group and the non-Covid ARVI dataset as a control group to train and test various machine learning (ML) models. Among these models, XGBoost and SVM-R demonstrated the highest performance, with accuracy values of 0.97 [0.93, 0.97] and 0.95 [0.95; 0.97], specificity values of 0.86 [0.71; 0.93] and 0.86 [0.79; 0.87], and sensitivity values of 0.984 [0.984; 1.000] and 1.000 [0.968; 1.000], respectively. When examining the Covid 2023 dataset, the Naive Bayes model achieved the highest classification accuracy of 43%, while XGBoost and SVM-R achieved accuracies of 14%. For the healthy control dataset, the accuracy of the models ranged from 0.27 [0.24; 0.32] for k-nearest neighbors to 0.44 [0.41; 0.45] for the Support Vector Machine with a radial basis function kernel. Conclusion: Therefore, ML models trained on MALDI MS of nasopharyngeal swabs obtained from patients with Covid during the initial phase of the pandemic, as well as symptomatic non-Covid individuals, showed excellent classification performance, which aligns with the results of previous studies. However, when applied to swabs from healthy individuals and a limited sample of patients with Covid in the late phase of the pandemic, ML models exhibited lower classification accuracy.

Keywords: SARS-CoV-2, MALDI-TOF MS, ML models, nasopharyngeal swabs, classification

Procedia PDF Downloads 87