HYBRID PREDICTION AND PRECEDENT-BASED DECISION SUPPORT USING MAHALLA ACTIVITY INDICATORS

Abstract

Generalized assessments of mahalla activity provide a useful basis for predicting expert scores and selecting relevant management recommendations from previously considered cases. This paper develops a hybrid workflow that integrates component assessment, supervised regression, and precedentbased decision support. Linear regression, second-degree polynomial regression, and a compact 3–2–1 multilayer perceptron are specified within a common training and evaluation protocol. Sequential parameter updates are combined with epoch-level validation and restoration of the best-performing checkpoint, contributing to stable and consistent model performance. At the second hybrid stage, the predicted score is linked to compatible historical cases, enabling the selection and transparent justification of appropriate management recommendations. The analysis distinguishes cross-sectional expert-score prediction from temporal forecasting and evaluates both numerical prediction accuracy and the relevance of retrieved recommendations. Structural results include parameter counts, bounded-output requirements, and an illustrative retrieval example. Historical findings complement the methodological analysis, while the numerical illustration demonstrates the practical implementation of the proposed approach. The developed framework provides a reproducible and systematic basis for evaluating prediction and decision-support quality and can support the further development of data-driven mahalla monitoring and management practices.