Concerning Internet of Things (IoT) systems, ensuring fault tolerance can be a challenging problem due to their complex, dynamic, and heterogeneous nature. This research presents a solution that integrates deep learning with Multi-Criteria Decision Analysis (MCDA) to enhance fault tolerance capabilities. By employing artificial intelligence techniques for real-time analysis, factors such as node security—monitored via Intrusion Detection Systems (IDS) using Long Short-Term Memory (LSTM) algorithms—are evaluated at the initial level. Subsequently, latency, energy consumption, and packet loss are assessed using a Naive Bayes approach. This methodology improves anomaly detection and contributes to the stability and reliability of IoT networks, ultimately leading to more robust and dependable systems.