DEEP LEARNING INTEGRATION TO ENHANCE MALWARE IDENTIFICATION IN INTELLIGENT IoT DEVICES

The research project aims to analyze and classify network traffic from IoT devices to detect ever-expanding and sophisticated cyber-attacks. The goal is to achieve high detection accuracy while reducing the time required detecting these attacks. The experimental study was conducted in various settings across Bangladesh over three months, beginning in January 2024 and ending in March 2024. The study proposes a lightweight malware detection framework using LSTM to classify malware into benign and hazardous categories. Python was used for data analysis. The framework uses a multi-layered distributive approach to manage data generation, communication, and device transfer. The model uses raw data to predict malware and classifies it into benign and hazardous categories. The dataset includes 88 benign files and 46 malware files, and features are selected for improved accuracy and performance. We have used the features ranked the highest using the feature selection process known as particle swarm optimization (PSO). In order to produce results that are both efficient and precise, it has utilized feature filtration approaches to determine which aspects are significant. As per findings, the PSO-KNN, GA-KNN, and SVM have collectively attained an accuracy of 99.6%, 98.4%, and 99.08%, respectively. 93.16% is the accuracy of the logistic regression. In contrast to the other machine learning methods, fuzzy c-mean has the least amount of accuracy. A hybrid LSTM and a hybrid CNN have both attained an accuracy of 99.6% and 99.2%. Deep learning classifiers are particularly ideal for identifying the ever-evolving sophisticated cyber threats. The time complexity is therefore calculated and compared with the time complexity of other classifiers that has been experimented with. The Internet of Things (IoT) has revolutionized daily life, but it also exposes devices to cyberattacks. A research study proposes a hybrid deep learning approach, CNN-CNN, with 99% detection accuracy. However, limitations include static malware detection and insufficient real-time testing.