DEEP LEARNING APPROACHES FOR SECURITY THREAT DETECTION AND MITIGATION IN INTERNET OF THINGS ENVIRONMENTS

Abstract

Background: The high-speed development of the Internet of Things (IoT) has had an important impact on the contemporary digital ecosystem, as it allows to interconnected smart devices in healthcare, industry, transportation, and smart cities. Nonetheless, IoT, environments are extremely susceptible to cyber-attacks because of limited resources, heterogeneous architectures, and weak authentication systems. Objective: The paper presents a hybrid deep learning model that can effectively detect and mitigation of security threats in IoT environment. Methodology: An experimental research design was chosen as quantitative. It created and tested a hybrid CNN-LSTM model using the IoT-23 dataset. Performance was compared to the conventional machine learning algorithms (SVM, Random Forest) and single models of deep learning (CNN, LSTM). The metrics used in evaluation were Accuracy, Precision, Recall, F1-score, and ROC-AUC. Results: The proposed CNN-LSTM model reached an accuracy of 98.4%, which was higher than comparative models. It showed better recall and F1-score in the detection of botnet, DDoS and malware-based IoT attacks. Conclusion: Hybrid deep learning architectures can improve the performance of IoT threat detectors to a considerable extent and provide real-time mitigation strategies.