Human Independent Real-Time Decision-Making Via Drift-Aware Adaptive Random Forest Ensembles

Real-time decision-making systems increasingly operate in environments characterized by continuous data streams, concept drift, and the need for immediate, autonomous responses without human intervention. Traditional batch-trained machine learning models struggle in such settings due to delayed retraining and limited adaptability. This paper proposes a novel, fully autonomous decision framework that employs Adaptive Random Forests (ARF) for real-time intelligent decision-making. The framework integrates online learning, drift-aware ensemble adaptation, and self-regulating decision logic to enable continuous learning and action without manual supervision. Unlike existing approaches that rely on periodic human-in-the-loop updates, the proposed system autonomously detects data distribution changes, reconfigures its ensemble structure, and issues decisions in real time. Experimental analysis on streaming benchmarks and simulated real-world scenarios demonstrates improved adaptability, stability, and decision latency compared to conventional online and batch learning models. The proposed approach is suitable for safety-critical and high-throughput applications such as smart infrastructure, cyber-physical systems, and autonomous monitoring platforms.