NeuroMetaQDS: An Interpretable Rule Based Quantized Decision System for Robust Regression Under Concept Drift

Real-world regression problems such as electricity price forecasting are often affected by concept drift, where the underlying data distribution changes over time and degrades the performance of static models. This paper proposes NeuroMetaQDS, an interpretable rule-based quantized decision system designed to remain robust under severe distribution shift. NeuroMetaQDS extracts a compact set of symbolic rules from a shallow decision tree, associates each rule with quantized weights, and refines them via a metalearning loop that runs over multiple meta-cycles. In each meta-cycle, the system adjusts rule thresholds and weights using simple, discrete updates driven by prediction error, while logging rule usage for interpretability. In a two-phase electricity market benchmark with strong drift between phases, NeuroMetaQDS achieves substantially lower Phase-2 error than Random Forest, XGBoost, and a sliding-window Adaptive Random Forest, while retaining competitive Phase-1 performance. These results show that meta-level rule adaptation can provide robust regression under concept drift without sacrificing model interpretability.