<![CDATA[The quality of clean and safe drinking water is crucial for public health; however, environmental pollution from industrial waste, domestic waste, and urbanization has significantly deteriorated water quality. Manual methods for water quality analysis, such as the Water Quality Index (WQI) and STORET, have limitations in efficiency and accuracy. Therefore, this study proposes a machine learning-based classification system to determine the potability of drinking water more accurately and efficiently. The Water Potability dataset from Kaggle, consisting of 3,276 samples with nine key parameters, was used in this research. Initial analysis showed that most features had a nearly normal distribution, although some variables, such as Solids and Conductivity, exhibited right-skewness due to extreme values. Correlation analysis revealed no significant linear relationships between water quality parameters. The preprocessing stage included missing data imputation using the mean method, normalization, feature engineering, and oversampling with SMOTE to address class imbalance. The machine learning models used in this study include LightGBM, Random Forest, XGBoost, and CatBoost, with model optimization performed using Bayesian Search CV, which improved performance, particularly for Random Forest. Experimental results showed that the optimized Random Forest model achieved the best performance with an accuracy of 85.38%, precision of 85.86%, recall of 85.38%, and an F1-score of 85.37%. However, some misclassifications remained, especially in detecting potable water samples, indicating that ensemble learning methods can be effectively used to evaluate drinking water potability.]]>
Copyrights © 2025
