Computational Intelligence and Network Systems

Video recommendation ensures that viewers get content more relevant to their choices and taste. With the aggregation of a diverse variety of content on video streaming platforms, there is a need to improve the existing recommendation model to increase accuracy in predicting the best content analogous to past choices of the user. The model proposed in this paper employs semantic similarity to generate recommendations based on metadata and annotations. The semantic inference-based model ensures the incorporation of cognitive knowledge of user inputs and past activities, yielding better insights into user preferences. The model utilizes web-sourced metadata to augment its dataset with supplementary information. The metadata is classified using RNN and is stored in a semantic network for feature selection. Topic modeling on query terms allows the discovery of hidden topics in the user query and the use of a knowledge store like DBpedia allows for enriching the query terms based on the topics discovered. The video dataset is categorized using a combination of Random Forest and decision tree-bagging techniques, which are based on the extracted features of the videos. The proposed SemVidRec model is compared with other video recommendation models like VAVR, PVRRC, CCVR, etc. SemVidRec was discovered to outperform the baseline models based on several performance metrics, achieving an accuracy of 96.45%, precision of 95.43%, and an nDCG value of 0.97.

Cardiovascular disease diagnosis and prompt medical care depend critically on the classification of heart sounds. In recent years, deep learning-based approaches have shown promising results in automating the process of heart sound categorization. This paper proposes a model HeartBeatNet (an attention UNet-based system) for heart sound classification that demonstrates comparatively better performance. The proposed system combines the strengths of attention mechanisms and the UNet architecture to effectively capture relevant features and to make accurate predictions. The system is trained on the PhysioNet/Cinc 2016 dataset consisting of annotated heart sound recordings, which are first converted into Mel Spectrograms before feeding into the UNet based network. The results indicate that the proposed system achieves high accuracy of 95.14%, sensitivity of 90.00%, and specificity of around 96.72% to classify various heart sound abnormalities.

A detection system for the diagnosis of Schizophrenia using machine learning and deep learning techniques are used in this study. Schizophrenia is a brain disorder which can be identified by various symptoms. Most common symptoms of Schizophrenia are speech disorder, laughing without any reason, crying without any reason, poor memory, lack of motivation etc. EEG signals are collected from human brains by placing electrodes (metal discs) on the scalp using a device named Electroencephalogram. It measures electrical activity of the brain, and the data is represented in the form of EEG signals. EEG signals are mainly used to study various diseases of the human brain. EEG signals of 14 healthy persons and 14 Schizophrenia patients are used. One machine learning classification algorithm, i.e. logistic regression and two deep learning models, i.e. convolutional neural network (CNN), and combination of multiple layers of convolutional neural networks and gated recurrent unit (GRU) are used to analyze the signals. Manual features are extracted from EEG signals and then feed into logistic regression to classify the signals. Extraction of Mel Frequency Cepstral Coefficient (MFCC) feature is done. Deep learning models are used to classify the EEG signals.

The model proposes image recommendation method by generating tags which is the state of the art in the evolving web 3.0. Proposed model works on the principle of enrichment of queries through topic Modelling and standard knowledge repositories. Data set driven topic synthesis and metadata synthesis by classifying it using Bi-LSTM classifier is the basis for the model. Strong semantic similarity computation measures such as Piyanka index Lance and William index and adaptive pointwise mutual index measures are integrated into the model. An intermediate semantic network is formalized, and Optimization is achieved using the harmonic search algorithm. Proposed MLSM is best among the baseline models with Precision of 94.09% recall of 96.91%.

6th generation networks are expected to be there most probably in year 2030. It is believed that their existence will enable the technology of Internet of Everything. People are already using the 5G Networks and facing several issues in that due to which the 6G networks are very important to develop. 6G technology ensures to connect every device present in the world with the internet providing Quality of Service. The 6G networks are still in discussion stage and it is identified that there is need of tackling the several issues related to security and implementation to get better and secure services from 6G Network. A lot of research is being carried out to ensure the security and integrity of 6G Technology. Various issues need to be addressed before implementing 6G Networks. The security is the main aspect behind the 6G Networks so security must be ensured at each layer of network model. The protocol security is very crucial issue to be considered. In this article, the security is the major concern that has been discussed in detail. Very few researchers have done the work on 6G protocol security in last three years. So, in this article, the main focus is to explore the security protocols and the issues related with the protocols to provide 6G network security have been elaborated to provide the way to the researchers to explore and resolve the 6G Security protocols. The future issues and challenges that need to be focused have also been discussed in this work.

Nowadays, the acquisition of deep knowledge is carried out in many fields like tracking objects from image/video, measuring position, acquiring textual and image content, visual value detection, and recognizing hand gestures. Various deep learning models are available based on classification and regression like Convolutional Neural Networks (CNN), decision trees, linear regression which comes under supervised and Self Organizing Map (SOM), Boltzmann Machines, and Autoencoders which comes under unsupervised models based on clustering. For image classification, one of the well-performing model is CNN. It provides good performance when compared to other deep learning models. In this paper, we propose an efficient method that will analyze sentiment from video. This is carried out by using Convolutional Neural Networks (CNN).

The findings are also compared with various well-known deep learning approaches, and the results obtained by the proposed method are higher when compared to existing ones. The proposed methodology gives an accuracy level of 92%. The proposed methodology can be used in various applications such as audio/video data sentiment analysis, monitoring social media, and customer feedback analysis, in the education field to draw student’s opinions.

Securing the Internet of Things-based environment is a top priority for consumers, businesses, and governments. There are billions of devices connecting and sharing data; an attack might cost billions of dollars. As a result, it’s important to protect the IoT network from external and internal threats. There is no way to guarantee that all vulnerabilities will be fixed with a single solution or that no additional flaws will be discovered. This paper proposes a deep learning-based solution to detect network intrusion in an IoT network to better prepare for network attacks. The proposed solution achieves the optimal tradeoff between accuracy and model weightage and ensures it is well-suited for resource-constrained IoT devices. The proposed solution uses a reduced data set for training produced by incremental PCA with LSTM, GRU, and BiLSTM. The proposed solution reduced the training time significantly while retaining the accuracy of 98.17% with GRU, 98.12% with LSTM, and 98.23% with BiLSTM, and the results show that the proposed model has better performance in training the model for detecting network intrusion in an IoT network.

Proteins play an important role in the human body and in plants. A lack of expertise in protein labeling in plants can make it extremely difficult to characterize and comprehend the precise roles and activities of different proteins. Furthermore, it restricts development in fields like biotechnology, disease resistance, and crop enhancement. The presented project focuses on plant protein classification, aiming to overcome the challenges arising from limited protein labeling knowledge. Advanced machine learning techniques, including various classification algorithms such as Logistic Regression, Decision Tree, K-nearest neighbors (KNN), Support Vector Machines (SVM), Random Forest (RF), Multinomial Naive Bayes (NB), AdaBoost, and XGBoost, are employed to accurately classify protein sequences into their respective families. This classification approach provides valuable insights into the functions and roles of proteins within plants, ultimately advancing our understanding of plant biology. This attempt offers new possibilities for advancement in critical sectors such as agriculture, drug discovery, and genomic research by eliminating the limitations associated with limited protein labeling knowledge.

Globally, Neurological disorders are a major health concern affecting a population of billions worldwide. There’s a need for accurate and timely diagnosis of brain disorders to improve patient outcomes and revolutionize the field of medicine with the help of technology. For this, the integration of deep learning models with MRI (structural and functional) images presents a promising approach for the detection of brain disorders like Alzheimer’s disease. Our Research aims to develop and evaluate deep learning models for detecting Alzheimer’s disease using the Oasis dataset, a popularly used data set of neuroimaging and processed imaging data, for brain images of Alzheimer patients. There were 2 types of images i.e. the Raw and FSL-SEG (preprocessed) gifs. The models were developed using multiple Convolution layers and a Non-linear activation function (Sigmoid) for binary classification. Early stopping on loss helped prevent overfitting, and a batch size of 75 was used for faster convergence. We generated an accuracy of 90% on the FSL-SEG MRI images whereas the RAW images resulted in an accuracy of 83%. With a value of 0.79 in Area Under the Curve, The CDR (Clinical Dementia Rating) as well as MMSE (Mini Mental State Examination) were main factors which interlinked the images with occurence of Alzheimer.

Natural disasters have dire consequences for communities, leading to loss of life, property destruction and environmental devastation. Effective disaster response necessitates prompt and coordinated actions. Social media platforms, particularly Twitter, have emerged as invaluable assets in disaster management. With an enormous daily influx of tweets, Twitter data presents an opportunity to gain valuable insights for tracking and responding to disasters. However, sifting through the vast volume of regular content to identify relevant tweets poses a significant challenge. Furthermore, the global nature of Twitter introduces an added layer of complexity with tweets in different languages. Recent advancements in deep learning techniques provide promising solutions for addressing this challenge, enabling the identification of disaster-related information from multilingual tweets. This research proposes a comprehensive approach that leverages machine learning and deep learning models to accurately classify disaster-related tweets in multiple languages, including English, Hindi, and Bengali. The study evaluates the performance of seven Machine Learning classifiers, including Naive Bayes, Logistic Regression, Random Forest, Support Vector Machine, K-Nearest Neighbors, Gradient Boosting, Decision Tree and few Deep Learning models such as LSTM, BiLSTM, BiLSTM with CNN, BERT and DistilBERT. After conducting a thorough evaluation of multiple models, it is evident that BERT and DistilBERT stand out as the top performers, consistently exhibiting exceptional accuracy and delivering consistent results across diverse language contexts.

Mitosis segmentation plays a vital role in early cancer detection, facilitating the accurate identification of dividing cells in histopathology images. Manual mitosis counting is time-consuming and subjective, prompting the need for automated approaches to improve efficiency and accuracy. In this study, we have developed a transformer-based U-Net model that combines the effectiveness of transformers which were originally designed for natural language processing (NLP) tasks, with the efficiency of the U-Net architecture to effectively capture both high-level and low-level features in histopathology images. We train and evaluate the model on the GZMH dataset and compare its performance against other deep models such as U-Net, U-Net++ and Mobilenetv2-based U-Net. The results demonstrate that transformer-based U-Net model is better in terms of accuracy, recall, precision, F1-score and Dice coefficient. This study represents a significant advancement in mitosis segmentation, contributing to improved cancer detection and prognosis.