Journal of Information Systems and Telecommunication (JIST) : 2 (46)

Open Access Article

1 - FLHB-AC: Federated Learning History-Based Access Control Using Deep Neural Networks in Healthcare System
Nasibeh Mohammadi Afshin Rezakhani Hamid Haj Seyyed Javadi Parvaneh asghari

Giving access permission based on histories of access is now one of the security needs in healthcare systems. However, current access control systems are unable to review all access histories online to provide access permission. As a result, this study first proposes a More

Giving access permission based on histories of access is now one of the security needs in healthcare systems. However, current access control systems are unable to review all access histories online to provide access permission. As a result, this study first proposes a method to perform access control in healthcare systems in real-time based on access histories and the decision of the suggested intelligent module. The data is used to train the intelligent module using the LSTM time series machine learning model. Medical data, on the other hand, cannot be obtained from separate systems and trained using different machine-learning models due to the sensitivity and privacy of medical records. As a result, the suggested solution employs the federated learning architecture, which remotely performs machine learning algorithms on healthcare systems and aggregates the knowledge gathered in the servers in the second phase. Based on the experiences of all healthcare systems, the servers communicate the learning aggregation back to the systems to control access to resources. The experimental results reveal that the accuracy of history-based access control in local healthcare systems before the application of the suggested method is lower than the accuracy of the access control in these systems after aggregating training with federated learning architecture. Manuscript profile

Open Access Article

2 - An Acoustic Echo Canceller using Moving Window to Track Energy Variations of Double-Talk-Detector
Mouldi MAKDIR Mourad BENZIANE Mohamed BOUAMAR

As a fundamental device in acoustic echo cancellation (AEC) systems, the echo canceller based on adaptive filters relies on the adaptive approximation of the echo-path. However, the adaptive filter must face the risk of divergence during the double-talk periods when the More

As a fundamental device in acoustic echo cancellation (AEC) systems, the echo canceller based on adaptive filters relies on the adaptive approximation of the echo-path. However, the adaptive filter must face the risk of divergence during the double-talk periods when the near-end is present. To solve this problem, the double-talk-detector (DTD) is often used to detect the double-talk periods and prevent the echo canceller from being disturbed by the other end of the speaker’s signal. In this paper, we propose a DTD based on a new method that can detect quickly and track accurately double-talk periods. It is based on the sum of energies of the estimated echo and the microphone signals which is continuously compared to the error energy. A window that moves with time and tracks energy variations of the different input signals of the DTD represents a fundamental feature of the proposed method compared to several other methods based on correlation. The goal is to outperform conventional normalized cross-correlation (NCC) methods which are well-known in terms of small steady-state misalignment and stability of decision variable. In this work, the normalized least mean squares (NLMS) algorithm is used to update the filter coefficients along speech signals which are taken from the NOIZEUS database. Efficiency of the proposed method is particularly compared to the conventional Geigel algorithm and normalized cross-correlation method (NCC) that depends on the cross-correlation between the microphone signal and the error signal of AEC. Performance evaluation is confirmed by computer simulation. Manuscript profile

Open Access Article

3 - An Aspect-Level Sentiment Analysis Based on LDA Topic Modeling
Sina Dami Ramin Alimardani

Sentiment analysis is a process through which the beliefs, sentiments, allusions, behaviors, and tendencies in a written language are analyzed using Natural Language Processing (NLP) techniques. This process essentially comprises of discovering and understanding people' More

Sentiment analysis is a process through which the beliefs, sentiments, allusions, behaviors, and tendencies in a written language are analyzed using Natural Language Processing (NLP) techniques. This process essentially comprises of discovering and understanding people's positive or negative sentiments regarding a product or entity in the text. The increased significance of sentiments analysis has coincided with the growth in social media such as surveys, blogs, Twitter, etc. The present study takes advantage of the topic modeling approach based on latent Dirichlet allocation (LDA) to extract and represent the thematic features as well as a support vector machine (SVM) to classify and analyze sentiments at the aspect level. LDA seeks to extract latent topics by observing all the texts, which is accomplished by assigning the probability of each word being attributed to each topic. The important features that represent the thematic aspect of the text are extracted and fed to a support vector machine for classification through this approach. SVM is an extremely powerful classification algorithm that provides the possibility to separate complex data from one another accurately by mapping the data to a space with much larger aspects and creating an optimal hyperplane. Empirical data on real datasets indicate that the proposed model is promising and performs better compared to the baseline methods in terms of precision (with 89.78% on average), recall (with 78.92% on average), and F-measure (with 83.50% on average). Manuscript profile

Open Access Article

4 - A Comparison Analysis of Conventional Classifiers and Deep Learning Model for Activity Recognition in Smart Homes based on Multi-label Classification
John Kasubi Manjaiah D. Huchaiah Ibrahim Gad Mohammad Kazim Hooshmand

Activity Recognition is essential for exploring the various activities that humans engage in within Smart Homes in the presence of multiple sensors as residents interact with household appliances. Smart homes use intelligent IoT devices linked to residents' homes to tra More

Activity Recognition is essential for exploring the various activities that humans engage in within Smart Homes in the presence of multiple sensors as residents interact with household appliances. Smart homes use intelligent IoT devices linked to residents' homes to track changes in human behavior as the humans interact with the home's equipment, which may improve healthcare and security issues for the residents. This study presents a research work that compares conventional classifiers such as DT, LDA, Adaboost, GB, XGBoost, MPL, KNN, and DL, focusing on recognizing human activities in Smart Homes using Activity Recognizing Ambient Sensing (ARAS). The experimental results demonstrated that DL Model outperformed with excellent accuracy compared to conventional classifiers in recognizing human activities in Smart Homes. This work proves that DL Models perform best in analyzing ARAS datasets compared to traditional machine learning algorithms. Manuscript profile

Open Access Article

5 - Designing a Semi-Intelligent Crawler for Creating a Persian Question Answering Corpus Called Popfa
Hadi Sharifian Nasim Tohidi Chitra Dadkhah

Question answering in natural language processing is an interesting field for researchers to examine their ability in solving the tough Alan Turing test. Every day computer scientists are trying hard to develop and promote question answering systems in various natural l More

Question answering in natural language processing is an interesting field for researchers to examine their ability in solving the tough Alan Turing test. Every day computer scientists are trying hard to develop and promote question answering systems in various natural languages, especially English. However, in Persian, it is not easy to advance these systems. The main problem is related to low resources and not enough corpora in this language. Thus, in this paper, a Persian question answering text corpus is created, which covers a wide range of religious, midwifery, and issues related to youth marriage topics and question types commonly encountered in Persian language usage. In this regard, the most important challenge was introducing a method for data gathering in Persian as well as facilitating and expanding the data gathering process. Though, SIC (Semi-Intelligent Crawler) is proposed as a solution that can overcome the challenge and find a way to crawl the Persian websites, gather text and finally import it to a database. The outcome of this research is a corpus called Popfa, which stands for POrsesh Pasokh (question answering) in FArsi. This corpus contains more than 53,000 standard questions and answers. Besides, it has been evaluated with standard approaches. All the questions in Popfa are answered by specialists in two general topics: religious and medical questions. Therefore, researchers can now use this corpus for doing research on Persian question answering. Manuscript profile

Open Access Article

6 - Whispered Speech Emotion Recognition with Gender Detection using BiLSTM and DCNN
Aniruddha Mohanty Ravindranath C. Cherukuri

Emotions are human mental states at a particular instance in time concerning one’s circumstances, mood, and relationships with others. Identifying emotions from the whispered speech is complicated as the conversation might be confidential. The representation of the spee More

Emotions are human mental states at a particular instance in time concerning one’s circumstances, mood, and relationships with others. Identifying emotions from the whispered speech is complicated as the conversation might be confidential. The representation of the speech relies on the magnitude of its information. Whispered speech is intelligible, a low-intensity signal, and varies from normal speech. Emotion identification is quite tricky from whispered speech. Both prosodic and spectral speech features help to identify emotions. The emotion identification in a whispered speech happens using prosodic speech features such as zero-crossing rate (ZCR), pitch, and spectral features that include spectral centroid, chroma STFT, Mel scale spectrogram, Mel-frequency cepstral coefficient (MFCC), Shifted Delta Cepstrum (SDC), and Spectral Flux. There are two parts to the proposed implementation. Bidirectional Long Short-Term Memory (BiLSTM) helps to identify the gender from the speech sample in the first step with SDC and pitch. The Deep Convolutional Neural Network (DCNN) model helps to identify the emotions in the second step. This implementation is evaluated with the help of wTIMIT data corpus and gives 98.54% accuracy. Emotions have a dynamic effect on genders, so this implementation performs better than traditional approaches. This approach helps to design online learning management systems, different applications for mobile devices, checking cyber-criminal activities, emotion detection for older people, automatic speaker identification and authentication, forensics, and surveillance. Manuscript profile

Open Access Article

7 - Ensemble learning of daboosting based on deep weighting for classification of hand-written numbers in Persian
amir asil hamed Alipour Shahram mojtahedzadeh hasan Asil

Today, the hand-written data volume is huge, which prohibits these data from being manually converted into electronic files. During the past years, different types of solutions were developed to convert machine learning-based handwritten data. Each method classifies or More

Today, the hand-written data volume is huge, which prohibits these data from being manually converted into electronic files. During the past years, different types of solutions were developed to convert machine learning-based handwritten data. Each method classifies or clusters the data according to the data type and application. In the present paper, a new approach is presented based on compound methods and deep learning for the classification of Persian handwritten data, where a deeper investigation is made of the data in basic learning by combining the Ada boosting and convolution. The present study aims at providing a new technique for classification of the images of handwritten Persian numbers. The structure of this technique is founded on Ada Boosting, which in turn, is based on weak learning. This technique improves learning by iteration of the weak learning processes and updating weights. In the meantime, the proposed method tried to employ stronger learners and present a stronger algorithm by combining these strong learners. The method was assessed on the standard Hoda dataset containing 60000 training data. The results show that the proposed method has a lower error rate than the previous methods by more than 1%. In the future, by developing basic learner, new mechanisms can be provided to improve the results by new types of learning. – Today, the hand-written data volume is huge, which prohibits these data from being manually converted into electronic files. During the past years, different types of solutions were developed to convert machine learning-based handwritten data. Each method classifies or clusters the data according to the data type and application. In the present paper, a new approach is presented based on compound methods and deep learning for the classification of Persian handwritten data, where a deeper investigation is made of the data in basic learning by combining the Ada boosting and convolution. The method was assessed on the standard Hoda dataset containing 60000 training data. The results showed that the error rate of the method has decreased by more than 1% compared to the previous methods. Manuscript profile

XML

List of Articles