A Modification of the K-Nearest Neighbor Algorithm in Assessment of Water Potability

Abstract

Water potability is a crucial necessity for public health, as access to clean and safe drinking water is vital for the prevention of waterborne diseases and the promotion of overall well-being. Contaminated water poses significant health hazards, including gastrointestinal infections, chronic diseases, and potential outbreaks of life-threatening ailments like cholera. Dependable evaluation techniques are essential for detecting hazardous water sources and facilitating prompt action to reduce hazards. In recent years, machine learning techniques have been versatile in solving classification problems as they can analyze and discover hidden patterns in the datasets which can possibly be too complex for human minds. In this study, we applied two such techniques called logistic regression and k-nearest neighbors for predicting the potability of a water body and attempted a modification of one of those two methods. The objective is to evaluate the two models by testing their accuracies and propose a new model which is more advanced at predicting accurately than the previous models. A dataset composed of 9 features of a water body is used to examine the efficiency of the models in assessing water quality. By presenting a detailed comparison of the methods and the results, we unlock a path for more modification in the future with the aim of further enhancing the performance and accuracy of the model.