Enhancing water potability prediction: Tuned vs default cat boost models

Water is a very important source for every living being on the earth. In such a scenario, access to clean and hygienic potable water remains a challenge in many parts of the world. Conventional water quality testing methods are expensive, and they need expertise. This challenge limits the scalability to test the water quality in environmental monitoring applications. To meet this challenge, a new attempt has been made by integrating the machine learning methods with chemical parameters to estimate the suitability of water for drinking purposes. The General and Tuned Cat Boost technique has been applied to the dataset, which contains the various physicochemical indicators. Various statistical parameters, such as accuracy, precision, recall, and F1-score, but also interpret their behaviour via confusion matrices, ROC curves, precision-recall plots, residual graphs, and feature importance scores, were evaluated for both models. On analysing these parameters, it was observed that the tuned model showed only marginal improvement over the general Cat Boost model. This paper concludes the observations of this study in a comprehensive framework that integrates performance, explainability, and model effectiveness for AI-empowered environmental intelligence systems.