Problem
As drinkable water becomes an increasingly scarce resource, quickly identifying whether a source is safe for human consumption matters, ideally from a small set of easily measurable variables.
Approach
Built classification models (Random Forest, Boosting) on 10 water-quality variables across a dataset covering 3,276 different water bodies. The potable/non-potable classes were imbalanced, so SMOTE was used to oversample the minority class and improve reliability.
Result
A working classifier for water potability, with the class-imbalance handling specifically called out as a deliberate technique to improve reliability rather than just reporting a misleading overall accuracy.