Journal of Computer and Knowledge Engineering
Volume:5 Issue: 2, Summer-Autumn 2022

With the aid of intelligent system approaches, the present study aimed at extracting and investigating effective features for detecting Attention-Deficit/Hyperactivity Disorder (ADHD) in children. With this end in view, 103 children, aged from 6 to 10, were recruited for this study, among which 49 cases were assigned to the treatment group (ADHD children) and the remaining 54 cases to the control group (healthy children). The disorder diagnosis was performed using the well-known, relevant psychological questionnaires and clinical interviews with expert psychologists. Data collection consisted of EEG signals in eyes open and eyes closed states, as well as GO/NOGO task for about 3 hours for every participant. The extracted features consisted of the amplitudes and latency in Event-Related Potential (ERP) and the power spectrum in the sleep mode signals. Approximately 826 features of 19 channels were extracted in the standard 10-20 system and different task conditions. A set of features were selected with the aid of the feature selection methods, and then the selected features were analyzed by neuroscientists, and the irrelevant ones were removed. Next, the classification methods and their performance evaluation were applied. Finally, the best results in terms of the corresponding feature vector and classification method were presented. The healthy and ADHD groups were classified with 75.8% accuracy using the Support Vector Machine (SVM) method. The results showed that the use of selection of effective features with the aid of intelligent system techniques under the supervision of experts leads us to reach robust biomarkers in the detection of disorders.

Representation learning on a knowledge graph aims to capture patterns in the knowledge graph as low-dimensional dense distributed representation vectors in the continuous semantic space, which is a powerful technique for predicting missing links in knowledge bases. The problem of knowledge base completion can be viewed as predicting new triples based on the existing ones. One of the prominent approaches in knowledge base completion is the embedding model. Currently, the majority of existing knowledge graph embedding models cannot deal with unbalanced entities and relations. In this paper, a new embedding model is proposed, with a general solution instead of using the additional corpus. First, a triple-based neural network is presented to maximize the likelihood of the knowledge bases finding a low-dimensional embedding space. Second, two procedures to generate positive triples are proposed. They produce positive triples and add them to the training data. The policies can capture rare triples, and simultaneously remain efficient to compute. Experiments show that the embedded model proposed in this paper has superior performance.

Social networks have become a central part of our lives these days and have real effects on the world's events. However, social networks greatly boost spreading misinformation and rumors that are becoming more and more dangerous each day. As fighting rumors first requires detecting them, several researchers tried to propose novel approaches for automatic early detection of rumors. However, most of them rely on handcrafted content features which makes them prone to deception and threats the adaptability of the model. Furthermore, a great deal of work have concentrated on event-level rumor detection while it faces early detection with serious challenges. There are also deficiencies in proposed methods in terms of time and resource complexity. This study proposes a deep learning approach to automate the detection of rumors on Twitter. The proposed method relies on automatically extracted features through word and sentence embeddings along with profile and network-based features. It then uses Recurrent Neural Networks (RNN) leveraging Gated Recurrent Units (GRU) for detecting the veracity of a tweet. The proposed method also improves time efficiency. The achieved experimental evaluation results on RumorEval2019 dataset demonstrate that the proposed method outperforms other rival models on the same dataset in terms of both performance and time complexity. By the way, the proposed method is more resilient to deception by avoiding the use of handcrafted content features and leveraging features that are out of the control of the user.

For more comprehensive security of a computer network as well as the use of firewall and anti-virus security equipment, intrusion detection systems (IDSs) are needed to detect the malicious activity of intruders. Therefore, the introduction of a high-precision intrusion detection system is critical for the network. Generally, the general framework of the proposed intrusion detection models is the use of text classification, and today deep neural networks (DNNs) are one of the top classifiers. A variety of DNN-based intrusion detection models have been proposed for software-defined networks (SDNs); however, these methods often report performance metrics solely on one well-known dataset. In this paper, we present a DNN-based IDS model with a 12-layer arrangement which works well on three datasets, namely, NSL-KDD, KDD99, and UNSW-NB15. The layered layout of the proposed model is considered the same for all the three datasets, which is one of the strengths of the proposed model. To evaluate the proposed solution, six other DNN-based IDS models have been designed. The values of the evaluation metrics, including accuracy, precision, recall, F-measure, and loss function, show the superiority of the proposed model over these six models. In addition, the proposed model is compared with several recent articles in this field, and the superiority of the proposed solution is shown.

Few-shot learning assumes that we have a very small dataset for each task and trains a model on the set of tasks. For real-world problems, however, the amount of available data is substantially much more; we call this a medium-shot setting, where the dataset often has several hundreds of data. Despite their high accuracy, deep neural networks have a drawback as they are black-box. Learning interpretable models has become more important over time. This study aims to obtain sample-based interpretability using the attention mechanism. The main idea is reducing the task training data into a small number of support vectors using sparse kernel methods, and the model then predicts the test data of the task based on these support vectors. We propose a sparse medium-shot learning algorithm based on a metric-based Bayesian meta-learning algorithm whose output is probabilistic. Sparsity, along with uncertainty, effectively plays a key role in interpreting the model's behavior. In our experiments, we show that the proposed method provides significant interpretability by selecting a small number of support vectors and, at the same time, has a competitive accuracy compared to other less interpretable methods.

Coronavirus 2019 (COVID-19), as a common infectious disease, is spreading rapidly and uncontrollably worldwide. Therefore, early detection of mortality considering the symptoms that appear in patients with Coronavirus is important. The main aim of this study is investigating the effect of data preprocessing methods on the efficiency of data mining approaches. In this study, we propose a hybrid method based on the Covid-19 dataset to predict the mortality of 1255 patients with coronavirus that has three main steps. In the first step, preprocessing methods such as imputing missing values, data balancing, normalization, and filter-based feature selection are used on raw data. Then the classification algorithms are applied to the data and finally, the evaluation is done. The results of the proposed method show its effectiveness in predicting mortality from coronavirus disease. Therefore, doctors and treatment staff can use this model to early diagnose of factors affecting the mortality of patients and with timely treatment, the mortality rate due to Covid-19 is reduced.