University of KashanSoft Computing Journal2322-370713120240822A review of machine learning algorithms to diagnose autism using the EEG signalA review of machine learning algorithms to diagnose autism using the EEG signal21911395510.22052/scj.2023.248522.1110FAZohrehSharifi MehrjardDepartment of Electrical Engineering, Technical and Vocational University, Tehran, Iran.HajarMomeniDepartment of Electrical Engineering, Ardakan University, Ardakan, Iran.HabibAdabi ArdekaniDepartment of Electrical Engineering, Yazd University, Yazd, Iran.Journal Article20221110Autism is a neurological and developmental disorder in which individuals often show limited symptoms or behaviors and are diagnosed based on a behavioral test. This neurological disorder is similar to other neurological disorders; therefore, diagnosing autism is a complicated task. If the disease is diagnosed at an early age, the severity of the disease and its complications will be reduced. In recent years, researchers have used electroencephalography (EEG) to diagnose autism. In the diagnosis of autism using EEG signals, machine learning algorithms have been used, and the human user has been able to diagnose this disease with a good percentage of accuracy by analyzing the extracted features. In this paper, the algorithms used in each of the articles have been reviewed, and the results obtained from the corresponding data review are the basis for further research. While examining the previously presented methods, the advantages and disadvantages of the methods are examined. The results show that the role of the methods used for pre-processing, extracting and selecting features from EEG images and the type of classifiers are effective factors in classification accuracy. Convolutional neural networks have been the most popular compared to other learning-based methods due to their ability to provide the best accuracy results. Autism is a neurological and developmental disorder in which individuals often show limited symptoms or behaviors and are diagnosed based on a behavioral test. This neurological disorder is similar to other neurological disorders; therefore, diagnosing autism is a complicated task. If the disease is diagnosed at an early age, the severity of the disease and its complications will be reduced. In recent years, researchers have used electroencephalography (EEG) to diagnose autism. In the diagnosis of autism using EEG signals, machine learning algorithms have been used, and the human user has been able to diagnose this disease with a good percentage of accuracy by analyzing the extracted features. In this paper, the algorithms used in each of the articles have been reviewed, and the results obtained from the corresponding data review are the basis for further research. While examining the previously presented methods, the advantages and disadvantages of the methods are examined. The results show that the role of the methods used for pre-processing, extracting and selecting features from EEG images and the type of classifiers are effective factors in classification accuracy. Convolutional neural networks have been the most popular compared to other learning-based methods due to their ability to provide the best accuracy results. https://scj.kashanu.ac.ir/article_113955_382846190c265e72d79da068c9df30bf.pdfUniversity of KashanSoft Computing Journal2322-370713120240822Using ensemble deep learning to improve the accuracy of CT-Scan lung image detection of COVID-19 patientsUsing ensemble deep learning to improve the accuracy of CT-Scan lung image detection of COVID-19 patients203911396110.22052/scj.2023.253142.1158FAMahdiSarchahiDepartment of Computer Engineering, Khavaran Institute of Higher Education, Mashhad, Iran.ElhamMahdipourDepartment of Computer Engineering, Khavaran Institute of Higher Education, Mashhad, Iran.Journal Article20230622Coronavirus disease 2019 or COVID-19 is an infectious disease caused by the severe acute respiratory syndrome virus (SARS-Cov-2) and has spread worldwide as an epidemic. After the rapid spread of the disease in 2019, the World Health Organization declared a public health emergency, and Human society saw a massive increase in deaths caused by its various mutations. Clinical symptoms include fever, cough, shortness of breath, and loss of smell. Fortunately, recently researchers have been able to achieve many successes in preventing the spread and speeding up its treatment by using different diagnostic methods. The aim of this research is to diagnose the disease of COVID-19 by processing CT scan images of people's lungs using ensemble deep learning techniques based on convolutional neural networks (CNN). In this regard, two data sets of CT scan images of people's lungs obtained from Kaggle and GitHub are used. The convolutional neural network architectures used in this research include VGG16, VGG19, Inception v3, ResNet50, DenseNet169, and CtNet10. In the first step, we added multiple dense (fully connected) layers to each of these models and evaluated their effect. Afterwards, to achieve higher accuracy and efficiency, the proposed ensemble method, which is a combination of VGG16, DenseNet169, and ResNet50 architectures, has been used. The experimental results show that the proposed ensemble method is able to achieve an accuracy of 98% to 100% on the investigated data set and significantly improve the performance of deep neural networks in multi-classification prediction tasks.Coronavirus disease 2019 or COVID-19 is an infectious disease caused by the severe acute respiratory syndrome virus (SARS-Cov-2) and has spread worldwide as an epidemic. After the rapid spread of the disease in 2019, the World Health Organization declared a public health emergency, and Human society saw a massive increase in deaths caused by its various mutations. Clinical symptoms include fever, cough, shortness of breath, and loss of smell. Fortunately, recently researchers have been able to achieve many successes in preventing the spread and speeding up its treatment by using different diagnostic methods. The aim of this research is to diagnose the disease of COVID-19 by processing CT scan images of people's lungs using ensemble deep learning techniques based on convolutional neural networks (CNN). In this regard, two data sets of CT scan images of people's lungs obtained from Kaggle and GitHub are used. The convolutional neural network architectures used in this research include VGG16, VGG19, Inception v3, ResNet50, DenseNet169, and CtNet10. In the first step, we added multiple dense (fully connected) layers to each of these models and evaluated their effect. Afterwards, to achieve higher accuracy and efficiency, the proposed ensemble method, which is a combination of VGG16, DenseNet169, and ResNet50 architectures, has been used. The experimental results show that the proposed ensemble method is able to achieve an accuracy of 98% to 100% on the investigated data set and significantly improve the performance of deep neural networks in multi-classification prediction tasks.https://scj.kashanu.ac.ir/article_113961_c9a2c339c42539493ebc668e9f0faa70.pdfUniversity of KashanSoft Computing Journal2322-370713120240822A survey on deep learning methods for text-based emotion classification: Advances, challenges, and opportunitiesA survey on deep learning methods for text-based emotion classification: Advances, challenges, and opportunities405711395810.22052/scj.2023.248812.1126FAMahdiRasouliDepartment of Engineering, University of Bojnord, Bojnord, IranVahidKianiDepartment of Engineering, University of Bojnord, Bojnord, IranJournal Article20230111Today, people on the web share their feelings and emotions with the help of various communication tools, one of the most common of which is the expression of feelings in textual content, such as social media posts, online store reviews, and user reviews. Emotion detection in text is a branch of sentiment analysis that aims to identify different types of human emotion in the text. This scientific field helps manufacturers and service providers to be aware of their weaknesses and strengths, and to provide better services to customers. In recent years, emotion recognition in text has become an attractive research field due to its wide applications in business, economics, politics, medicine, psychology, and sociology. In this article, the problem of emotion classification in text and its solution methods will be investigated with emphasis on deep learning. Also, a brief description of the latest deep learning solutions that have been used in recent years to classify emotion in text will be discussed. In addition, some labelled datasets, the most important open issues in emotion recognition, and future research directions will also be presented, which can be a good guide for new researchers in this field.Today, people on the web share their feelings and emotions with the help of various communication tools, one of the most common of which is the expression of feelings in textual content, such as social media posts, online store reviews, and user reviews. Emotion detection in text is a branch of sentiment analysis that aims to identify different types of human emotion in the text. This scientific field helps manufacturers and service providers to be aware of their weaknesses and strengths, and to provide better services to customers. In recent years, emotion recognition in text has become an attractive research field due to its wide applications in business, economics, politics, medicine, psychology, and sociology. In this article, the problem of emotion classification in text and its solution methods will be investigated with emphasis on deep learning. Also, a brief description of the latest deep learning solutions that have been used in recent years to classify emotion in text will be discussed. In addition, some labelled datasets, the most important open issues in emotion recognition, and future research directions will also be presented, which can be a good guide for new researchers in this field.https://scj.kashanu.ac.ir/article_113958_5fd13318424cee492e77cfa2d728604b.pdfUniversity of KashanSoft Computing Journal2322-370713120240822Localization of mobile targets in a wireless sensor network using Diffusion Least Mean Square algorithm based on Huber loss functionLocalization of mobile targets in a wireless sensor network using Diffusion Least Mean Square algorithm based on Huber loss function587511396310.22052/scj.2023.252719.1141FASoheilaAshkezariFaculty of Mathematical Sciences, Ferdowsi University of Mashhad, Mashhad, Iran.Department of Computer Engineering, Salman Institute of Higher Education, Mashhad, Iran.Mohammad-NaeemTeimooriDepartment of Computer Engineering, Salman Institute of Higher Education, Mashhad, Iran.Vahid-RezaSabzevariDepartment of Biomedical Engineering, Mashhad Branch, Islamic Azad University, Mashhad, Iran.Journal Article20230404Localization of mobile targets is one of the important topics in wireless sensor networks. The challenge lies in deploying techniques capable of estimating the subject's location amidst system noise, with minimal deviation from the actual location. In this paper, we propose a robust variant of the Diffusion Least Mean Square algorithm. This version involves distributing the estimation of the target's location across network nodes, facilitated by the pseudo-Huber loss function. Through this method, the accuracy of estimation in localization and tracking the target improves even in the presence of various noise types. The paper formulates target location using two criteria: received signal strength and signal propagation time, based on the proposed algorithm within an adaptive filter network. Experimental results highlight the algorithm's capability to enhance the accuracy of localization and tracking operations. This improvement remains consistent across wireless sensor network scenarios influenced by both Gaussian and non-Gaussian noises, with varying signal-to-noise ratios.Localization of mobile targets is one of the important topics in wireless sensor networks. The challenge lies in deploying techniques capable of estimating the subject's location amidst system noise, with minimal deviation from the actual location. In this paper, we propose a robust variant of the Diffusion Least Mean Square algorithm. This version involves distributing the estimation of the target's location across network nodes, facilitated by the pseudo-Huber loss function. Through this method, the accuracy of estimation in localization and tracking the target improves even in the presence of various noise types. The paper formulates target location using two criteria: received signal strength and signal propagation time, based on the proposed algorithm within an adaptive filter network. Experimental results highlight the algorithm's capability to enhance the accuracy of localization and tracking operations. This improvement remains consistent across wireless sensor network scenarios influenced by both Gaussian and non-Gaussian noises, with varying signal-to-noise ratios.https://scj.kashanu.ac.ir/article_113963_acf6e3e819955af1189e27f8bb12c51e.pdfUniversity of KashanSoft Computing Journal2322-370713120240822Designing a stable model for the Tehran subway train scheduling problem with a robust optimization approachDesigning a stable model for the Tehran subway train scheduling problem with a robust optimization approach769511425310.22052/scj.2024.248714.1120FADavoodJafariFaculty of Industrial Engineering, Islamic Azad University, Parand Branch, Tehran, Iran.MehranKhalajFaculty of Industrial Engineering, Islamic Azad University, Parand Branch, Tehran, Iran.PejmanSalehiFaculty of Industrial Engineering, Islamic Azad University, Parand Branch, Tehran, Iran.Journal Article20221219This study aims to present a robust optimization model for scheduling intra-city rail transport trains (metro) in Tehran. We present a new approach based on robust principles and concepts to tackle the challenge of timetable planning for Tehran's subway trains under uncertainty and disturbances. To this end, a simulation method is employed, considering buffer times and minimum train movement intervals. Therefore, this paper follows a linear programming framework to determine the optimal headway distances for Tehran's subway lines, with the goal of minimizing average waiting times at station platforms and enhancing train flow according to passenger demand. Using a general linear programming model, after defining the objective function and constraints—such as train stopping times at stations (nodes), route lengths (edges), station capacities, and safety margins in line with the overall rail network conditions—optimal points are identified through buffer times and simulation approaches, exemplified at Shahed station, and incorporated into the final timetable design. Additionally, this work assumes that parameters related to capacity and timing of train units are non-deterministic. At the end, the validity of the proposed model is evaluated using data from the southern part of line one of Tehran's subway, and the results are presented.This study aims to present a robust optimization model for scheduling intra-city rail transport trains (metro) in Tehran. We present a new approach based on robust principles and concepts to tackle the challenge of timetable planning for Tehran's subway trains under uncertainty and disturbances. To this end, a simulation method is employed, considering buffer times and minimum train movement intervals. Therefore, this paper follows a linear programming framework to determine the optimal headway distances for Tehran's subway lines, with the goal of minimizing average waiting times at station platforms and enhancing train flow according to passenger demand. Using a general linear programming model, after defining the objective function and constraints—such as train stopping times at stations (nodes), route lengths (edges), station capacities, and safety margins in line with the overall rail network conditions—optimal points are identified through buffer times and simulation approaches, exemplified at Shahed station, and incorporated into the final timetable design. Additionally, this work assumes that parameters related to capacity and timing of train units are non-deterministic. At the end, the validity of the proposed model is evaluated using data from the southern part of line one of Tehran's subway, and the results are presented.https://scj.kashanu.ac.ir/article_114253_5f7fdf27e7394a046ca198b1a6861120.pdfUniversity of KashanSoft Computing Journal2322-370713120240822Improve opinion mining by combining gray wolf algorithm and support vector machineImprove opinion mining by combining gray wolf algorithm and support vector machine9610711467310.22052/scj.2025.248730.1122FAFaribaSalahiDepartment of Industrial Management, Faculty of Management, Islamic Azad University, South Tehran Branch, Tehran, Iran.Journal Article20221225The advent of the web and its continuous growth have created a tremendous amount of user-generated information. Valuable subjective information is easy to find. Especially on social networks and e-commerce platforms that contain essential information. As a result, the field of belief mining has attracted considerable attention in recent years. New research papers are published every day in which various AI techniques are applied to various tasks and applications related to mining opinion. In this article, a new approach based on the support vector machine technique and the gray wolf optimization algorithm is proposed to improve the opinion mining process. Such that the gray wolf algorithm is used to determine the practical features in the belief mining process and enhances the performance of the support vector machine. This research showed that the proposed system could increase the accuracy and cover the error of the backup vector technique by selecting practical features. The proposed approach has been evaluated using three criteria: accuracy, recall, and precision. Precision for the first and second classes is 0.68 and 0.92%, respectively, and recall for the first and second classes equals 0.94 and 0.63% respectively, and accuracy was 0.77%. The results indicate that the proposed system of this research achieved favorable results in both categories.The advent of the web and its continuous growth have created a tremendous amount of user-generated information. Valuable subjective information is easy to find. Especially on social networks and e-commerce platforms that contain essential information. As a result, the field of belief mining has attracted considerable attention in recent years. New research papers are published every day in which various AI techniques are applied to various tasks and applications related to mining opinion. In this article, a new approach based on the support vector machine technique and the gray wolf optimization algorithm is proposed to improve the opinion mining process. Such that the gray wolf algorithm is used to determine the practical features in the belief mining process and enhances the performance of the support vector machine. This research showed that the proposed system could increase the accuracy and cover the error of the backup vector technique by selecting practical features. The proposed approach has been evaluated using three criteria: accuracy, recall, and precision. Precision for the first and second classes is 0.68 and 0.92%, respectively, and recall for the first and second classes equals 0.94 and 0.63% respectively, and accuracy was 0.77%. The results indicate that the proposed system of this research achieved favorable results in both categories.https://scj.kashanu.ac.ir/article_114673_ff5279c41bc0e9c3dd77b1ae684c10f3.pdfUniversity of KashanSoft Computing Journal2322-370713120240822Agent-based models and their application in investigating the risks of epidemic disease transmissionAgent-based models and their application in investigating the risks of epidemic disease transmission10812111426010.22052/scj.2024.253185.1165FAMahmoodDadkhahDepartment of Mathematics, Payame Noor University, Tehran, Iran.NazliBesharatiDepartment of Mathematics, Payame Noor University, Tehran, Iran.Journal Article20230705The outbreak of the COVID-19 pandemic in the world once again faced the governments and local officials with a big and unexpected challenge. Examining all aspects of this epidemic brought experts from different fields, especially mathematics, into the field of fighting this challenge. In the meantime, mathematical modeling of disease transmission has gained double importance due to its simple and understandable nature as well as its proven ability to help solve such issues. Different models presented for different diseases increase the power of investigation and prevention of dangerous epidemics. In this article, with the help of agent-based models, we have investigated the spread and transmission of epidemic diseases (especially in small environments). In the reviewed model, the spatio-temporal transfer process is also considered for each agent. Also, the decision of the investigated agents will be based on the rules prepared for them. To define the main social characteristics as well as the health conditions used during the interaction of the agents with each other, an individual profile is predicted for each agent in the model. Due to the good flexibility of the stated model, different numerical simulations have been implemented with it. From what was obtained in this research, it can be seen that in order to control and deal with epidemic diseases, especially in small environments, several factors must be taken seriously, of which, serious and severe restrictions on the movement of agents, their gatherings and the observance of health and protective instructions (for example, using a mask and quarantine agents) cab be pointed out. The outbreak of the COVID-19 pandemic in the world once again faced the governments and local officials with a big and unexpected challenge. Examining all aspects of this epidemic brought experts from different fields, especially mathematics, into the field of fighting this challenge. In the meantime, mathematical modeling of disease transmission has gained double importance due to its simple and understandable nature as well as its proven ability to help solve such issues. Different models presented for different diseases increase the power of investigation and prevention of dangerous epidemics. In this article, with the help of agent-based models, we have investigated the spread and transmission of epidemic diseases (especially in small environments). In the reviewed model, the spatio-temporal transfer process is also considered for each agent. Also, the decision of the investigated agents will be based on the rules prepared for them. To define the main social characteristics as well as the health conditions used during the interaction of the agents with each other, an individual profile is predicted for each agent in the model. Due to the good flexibility of the stated model, different numerical simulations have been implemented with it. From what was obtained in this research, it can be seen that in order to control and deal with epidemic diseases, especially in small environments, several factors must be taken seriously, of which, serious and severe restrictions on the movement of agents, their gatherings and the observance of health and protective instructions (for example, using a mask and quarantine agents) cab be pointed out. https://scj.kashanu.ac.ir/article_114260_957e33f23ff29462807a7125a2f4c4bd.pdfUniversity of KashanSoft Computing Journal2322-370713120240822Random walk on connected graphs and its application in electronic networksRandom walk on connected graphs and its application in electronic networks12215711425210.22052/scj.2024.248446.1106FAMehdiShamsDepartment of Statistics, Faculty of Mathematical Sciences, University of Kashan, Kashan, Iran.GholamrezaHesamianDepartment of Statistics, Faculty of Mathematics, Payame Noor University, Tehran, Iran.Journal Article20221016In this article, random walk parameters are analyzed. Then the relations of eigenvalues are examined. Meanwhile, a limit is set for the main parameters. Then, the application in electronic networks will be described. Also, applications in computer science are mentioned, especially in encryption management for the computing network. Physical achievements are used to obtain results in random walks. At the end, the applied algorithms of random walk and random walk sampling will be described.In this article, random walk parameters are analyzed. Then the relations of eigenvalues are examined. Meanwhile, a limit is set for the main parameters. Then, the application in electronic networks will be described. Also, applications in computer science are mentioned, especially in encryption management for the computing network. Physical achievements are used to obtain results in random walks. At the end, the applied algorithms of random walk and random walk sampling will be described.https://scj.kashanu.ac.ir/article_114252_88f1955fb84cade5dfb7835f2f3237d6.pdfUniversity of KashanSoft Computing Journal2322-370713120240822Face recognition with incomplete data by deep convolutional neural networkFace recognition with incomplete data by deep convolutional neural network15817111425410.22052/scj.2024.252789.1143FAFarnazHoseiniDepartment of Computer Engineering, Technical and Vocational University (TVU), Tehran, Iran.ElaheTabibzade LamarDepartment of Computer Engineering, Shahriar Institute of Higher Education, Astara, Iran.Seyed MehdiMirkazemi NiaragDepartment of Electronics, Islamic Azad University, Astara Branch, Astara, Iran.Journal Article20230412The human face is not a fixed entity influenced by various factors that give rise to different facial expressions. Face recognition algorithms encounter challenges such as inherent and random factors causing facial appearance variations, incomplete data in the database, database size, differences in image dimensions, and changes in facial expressions. Addressing these challenges can expand the application range of facial recognition techniques. In this study, we propose a method that utilizes a deep convolutional neural network to enhance face recognition in the presence of incomplete data. The proposed method consists of several distinct steps. Firstly, primary data is selected and extracted from the database, followed by preprocessing the information through filtering, histogram transformation, and edge detection. The output of this step serves as input to the bee optimization algorithm, which facilitates the selection of relevant features and optimizes them for recognition. Finally, a deep convolutional neural network is employed for face recognition, encompassing training and testing stages. We conducted simulations in the MATLAB environment to evaluate the proposed method and assess using accuracy, correctness, and criteria coverage criteria. The results demonstrated an accuracy of 96.11%, indicating improved face recognition compared to recent works and cost reductions in the overall recognition process.The human face is not a fixed entity influenced by various factors that give rise to different facial expressions. Face recognition algorithms encounter challenges such as inherent and random factors causing facial appearance variations, incomplete data in the database, database size, differences in image dimensions, and changes in facial expressions. Addressing these challenges can expand the application range of facial recognition techniques. In this study, we propose a method that utilizes a deep convolutional neural network to enhance face recognition in the presence of incomplete data. The proposed method consists of several distinct steps. Firstly, primary data is selected and extracted from the database, followed by preprocessing the information through filtering, histogram transformation, and edge detection. The output of this step serves as input to the bee optimization algorithm, which facilitates the selection of relevant features and optimizes them for recognition. Finally, a deep convolutional neural network is employed for face recognition, encompassing training and testing stages. We conducted simulations in the MATLAB environment to evaluate the proposed method and assess using accuracy, correctness, and criteria coverage criteria. The results demonstrated an accuracy of 96.11%, indicating improved face recognition compared to recent works and cost reductions in the overall recognition process.https://scj.kashanu.ac.ir/article_114254_8c2c1bcad07bdef193dde963f2af0256.pdfUniversity of KashanSoft Computing Journal2322-370713120240822Numerical simulation of water long waves modeled by nonlinear Boussinesq partial differential equation using a spectral approximationNumerical simulation of water long waves modeled by nonlinear Boussinesq partial differential equation using a spectral approximation17218111426410.22052/scj.2024.253459.1178FAHooriehFakhariDepartment of Applied Mathematics, Faculty of Mathematical Sciences, University of Kashan, Kashan, Iran.AkbarMohebbiDepartment of Applied Mathematics, Faculty of Mathematical Sciences, University of Kashan, Kashan, Iran.Journal Article20230822In this paper, the Galerkin method is proposed for the solution of the nonlinear Boussinesq partial differential equation describing water waves. The main idea is to use generalized Jacobi polynomials (GJPs) as basis functions to deal with spatial derivatives such that boundary conditions are satisfied. To avoid solving nonlinear equations, the Leap-frog and Crank-Nicolson method is proposed for time discretization of the equation. The error estimate of the proposed method is investigated and numerical results show the high accuracy and low CPU time of proposed method and confirmed the theoretical ones. Also, the obtained results show that the method is suitable for nonlinear and even-order partial differential equations. In this paper, the Galerkin method is proposed for the solution of the nonlinear Boussinesq partial differential equation describing water waves. The main idea is to use generalized Jacobi polynomials (GJPs) as basis functions to deal with spatial derivatives such that boundary conditions are satisfied. To avoid solving nonlinear equations, the Leap-frog and Crank-Nicolson method is proposed for time discretization of the equation. The error estimate of the proposed method is investigated and numerical results show the high accuracy and low CPU time of proposed method and confirmed the theoretical ones. Also, the obtained results show that the method is suitable for nonlinear and even-order partial differential equations. https://scj.kashanu.ac.ir/article_114264_22fb24fd8bf3f4c9c912f9d47e35e25f.pdf