Abstract

This study presents an AI-powered solution addressing three critical challenges in IoT security: Class imbalance in intrusion datasets, limited diversity in attack patterns, and the need for real-time analysis of network traffic. Our hybrid approach combines optimized machine learning models (Logistic Regression, Random Forest with One-vs-Rest strategy) with advanced deep learning architectures (FNN, LSTM, GRU) to handle complex multi-class classification of network intrusions. We developed an end-to-end detection system featuring a user-friendly Streamlit interface that supports PCAP file analysis, real-time traffic visualization, and interactive threat reporting. The system achieved 98.56% detection accuracy in our evaluations, significantly outperforming conventional methods. Practical implementation features include automated firewall configuration recommendations and dynamic IP blocking for high-risk threats. This research demonstrates how AI-driven security solutions can provide adaptive protection for IoT ecosystems, offering both technical innovation and practical utility.