Enhanced Fault and Attack Detection in IoT Systems with Naive Bayes and Long Short-Term Memory Models

As the number of IoT-dependent gadgets increases in the market, stringent security and fault detection solutions are required. The characteristics and variability of IoT systems may not fully capture classical detection techniques, leading to the use of more advanced machine learning-based techniques. This research employs a dual detection model consisting of Naive Bayes and Long Short-Term Memory (LSTM). The approach analyzes byte streams originating from IoT networks and device patterns, normalizes and encodes the data, and divides the classification process into two phases: screening through Naive Bayes and detailed analysis through LSTM. The proposed hybrid model achieved an overall fault detection accuracy of 97.3 % and an attack detection accuracy of 96.8 % for various IoT scenarios. The system achieved 95.2% precision and 94.8% recall, as well as a 2.1% false-positive rate. The results demonstrated that the proposed hybrid approach is 12 and 15 percent faster than the standalone implementations. In critical scenarios, particularly those where attackers employ multi-vector attacks against IoT systems and hard-to-detect faults in the communication and runtime environments, the proposed framework offers a viable solution to address these issues and challenges.