Machine Learning-based Performance Evaluation of XGBoost Compared with KNN and MLP for Local Water Quality Assessment
Abstract
This study seeks to create a smart water quality analysis system with machine learning to help accurately evaluate the state of the local water bodies and solve the issue of low prediction accuracy. Three machine learning models were tested in the given study. Group 1 used the K-Nearest Neighbour (KNN) algorithm to predict the water quality using standard physicochemical parameters. Group 2 used the Multilayer Perceptron (MLP) framework to enhance the learning ability by using the neural network-based classification. The proposed XGBoost model was applied in group 3, implying that it operates with already processed data of the water quality, including normalization, feature scaling, and balancing of the data. A dataset consisting of several samples of water was trained and utilized to test all the three models. The experimental evidence indicated that the XGBoost model performed better with an accuracy of 81.2, precision and recall values are greater to KNN and MLP models. This paper concludes that XGBoost is a more stable and reliable tool in the classification of water quality in the local water bodies than the traditional machine learning models. Its better performance, in turn, makes it applicable to real-world environmental monitoring applications, even in a resource-limited environment.