International Journal of Scientific Research and Engineering Development

Decision-support systems operating in non-stationary environments are susceptible to performance degradation due to distributional changes in input data. Conventional drift-aware approaches typically rely on purely statistical detectors, which may trigger unnecessary model updates in response to transient fluctuations or noise. This paper presents a context-aware adaptive prediction system that integrates statistical drift monitoring with external contextual metadata to regulate retraining decisions under evolving conditions. Drift alarms generated by statistical detectors are validated against documented contextual events, ensuring that adaptation is initiated only when distributional changes are operationally meaningful. The framework incorporates cost-sensitive retraining and adaptive decision thresholding to address class imbalance while maintaining stable predictive behavior. Experimental evaluation on longitudinal datasets demonstrates improved stability and reduced spurious retraining compared to purely statistical drift-based approaches. Student dropout prediction is used as a representative application case study to evaluate the proposed system under non-stationary conditions.

Chaitanya Mokkapati, Sayyed Nagulmeera, Shaik Aasha, M Dileep Kumar, "A Context-Validated Adaptive Prediction System for Decision Support Under Concept Drift " International Journal of Scientific Research and Engineering Development, V8(6): Page(3162-3171) Nov-Dec 2025. ISSN: 2581-7175. www.ijsred.com. Published by Scientific and Academic Research Publishing.