The Internet of Things (IoT) has transformed modern computing ecosystems by enabling large-scale data generation and real-time analytics. Traditional database optimizers and offline-trained predictive models often fail under sudden workload shifts and concept drift, resulting in degraded performance and high resource consumption. ALP-DB continuously learns from evolving IoT workloads, autonomously reconfigures database parameters, and ensures compliance with latency and energy constraints. Experimental evaluation using real-world and synthetic IoT workloads demonstrates that ALP-DB outperforms static, heuristic, and reinforcement learning-based baselines by reducing average query latency by up to 28%, tail latency by 35%, and energy consumption by 20%, while maintaining low reconfiguration overhead.