Semantic-aware reinforcement and ensemble learning for signal management and anomaly detection in IoT systems

The rapid growth of large-scale internet of things (IoT) networks poses critical challenges in maintaining reliable signal quality and timely anomaly detection under heterogeneous and dynamic environments. This paper presents a semantic-aware hybrid framework that integrates deep reinforcement learning (DRL) and random forest (RF) classification for intelligent signal optimization and anomaly detection. The framework leverages semantically enriched features-including mean Received Signal Strength Indicator RSSI, number of active base stations, and geospatial-temporal context-to capture both signal quality and environmental dynamics. A Deep Q-Network (DQN) agent learns optimal signal allocation strategies, while the RF classifier achieves high-accuracy prediction of connection states, and an isolation forest (IF) module identifies anomalies in real-time. Experimental evaluation on Sigfox and LoRaWAN datasets demonstrates that the DQN agent achieves a stable cumulative reward above 62,000, the RF classifier attains 99.98% accuracy, and the IF-based anomaly detection module achieves 99.97% accuracy with precision, recall, and F1-scores of 99.95%, successfully detecting 719 anomalous instances out of 14,377 records, validating the framework's effectiveness in practical IoT deployments. Compared with existing single-method approaches, the proposed framework unifies optimization, verification, and early-warning anomaly detection in a closed semantic-aware loop, providing robust, context-aware network management. These results highlight the framework's potential to enhance communication efficiency, maintain service reliability, and support adaptive operation in real-time heterogeneous IoT networks, demonstrating its suitability for resilient next-generation IoT and 6G-enabled systems.