Smart environments frequently experience fluctuations in demand, infrastructure usage, and service-level expectations, requiring adaptive systems capable of dynamic selfoptimization. This work proposes a self-adaptive framework that integrates real-time monitoring, anomaly prediction, and reinforcement learning (RL) to proactively reconfigure systems before degradation occurs. Unlike previous approaches limited to specific layers or metrics, the proposed solution supports diverse performance indicators-ranging from hardware and software to network and SLA levels-and offers an extended repertoire of adaptation actions, such as resource scaling, node restarts, and heuristic tuning. Its design leverages predictive alerts to reduce reaction delays and improve adaptation timing. Experimental validation demonstrated a mean adaptation time of 0.05 seconds, approximately 94% adaptation accuracy, 4% overhead, and 98% operational stability, confirming its effectiveness for real-time optimization in heterogeneous and dynamic environments.

--

Anomaly detection,Predictive analytics,Reinforcement learning,Self-adaptive systems,Smart environments

• Ciências da Computação e da Informação - Ciências Naturais
• Engenharia Eletrotécnica, Eletrónica e Informática - Engenharia e Tecnologia