Article’s

The Mahatma Gandhi National Rural Employment Guarantee Act (MGNREGA) represents a cornerstone of India’s social security infrastructure, disbursing substantial fiscal resources annually to sustain rural livelihoods. However, traditional management practices within public welfare administration remain largely reactive, relying heavily on historical baselines and manual planning paradigms that exhibit structural limitations in predicting localized demand spikes and dynamic climate disruptions. This paper introduces SchemeImpactNet, a datadriven predictive and optimization framework engineered to transition public welfare administration from reactive management to evidence-based proactive policy intervention. Leveraging an extensive administrative dataset comprising 7,758 districtyear records across 759 districts and 34 states/union territories from 2014–15 to 2024–25, we evaluate a series of rigorous predictive machine learning algorithms under a strictly leakfree temporal validation protocol. Gradient Boosting models achieve a cross-validated mean R2 of 0.9078 (excluding anomaly years), outperforming alternative parametric and non-parametric algorithms. Utilizing these localized predictions, we model a linear programming resource optimization engine using the PuLP framework to maximize public employment outcomes under fixed historical budgetary parameters. Experimental results establish that mathematically optimized resource reallocation across administrative subdivisions yields a simulated employment gain of 1,840.41 lakh person-days, representing a 6.38% net improvement in nationwide execution efficiency without increasing nominal public expenditure. The architecture is containerized and deployed via an open REST API and interactive Streamlit service, establishing reproducible methodology for public policy optimization. Index Terms—Public Policy Analytics, Predictive Modeling, Gradient Boosting, Resource Allocation Optimization, Linear Programming, MGNREGA.