Processed CPCB-NWMP 2024 Water Quality Dataset and Analysis Code for Explainable Ensemble Classification of Indian Lakes, Ponds, and Tanks
Description
This deposit provides the filtered, feature-engineered analytical dataset and complete analysis code accompanying the manuscript "Explainable Ensemble Learning for Multiclass Water Quality Classification and Geographical Generalisation Using India's Central Pollution Control Board Monitoring Data" (Neetha Alex &Dr R Brindha, submitted to Expert Systems With Applications). The dataset (processed_683station_dataset.csv) contains 683 monitoring stations retained from India's Central Pollution Control Board National Water Monitoring Programme (CPCB-NWMP) 2024 release, covering lakes, ponds, and tanks across multiple Indian states and union territories. Thirteen engineered features (eight parameter means, five within-year ranges) are derived from eight raw physicochemical parameters (temperature, dissolved oxygen, pH, electrical conductivity, biochemical oxygen demand, nitrate, faecal coliform, and total coliform), along with a four-class water-quality label (Excellent, Good, Poor, Very Poor) assigned by a precedence-ordered rule adapted from CPCB Designated Best Use criteria. The accompanying Python code reproduces every headline result in the manuscript: the eight-classifier baseline comparison and stacking-ensemble architecture, twelve-state Leave-One-State-Out geographical generalisation validation, a ten-seed-robust ablation quantifying target-construction dependence (the extent to which classification accuracy reflects the same variables used to construct the class labels), and statistical significance testing for the Telangana state-level outlier. This deposit does not include the raw, unedited CPCB-NWMP source file, which remains publicly available directly from CPCB (https://cpcb.nic.in/nwmp-data/) and is not redistributed here.