Machine learning modeling the classification task for water quality prediction

Water quality prediction is of great importance for ecosystem stability. We propose effective machine learning methodologies applied to two public water datasets. The proposed methodologies have two characteristics: (a) they offer an efficient approach to accounting for unbalanced data and (b) they skillfully combine classical machine learning models such as Random Forest, Decision Tree, K-Nearest Neighbors, XGBoost, and Support Vector Machines. Methodology 1 overcomes the noticeable class imbalance in the 'Water Quality' dataset by implementing SMOTE with a “sampling_strategy=0.50”. This approach in significant improvements in model accuracy and a reduction in processing time. Methodology 1 achieves the highest accuracy (98.92% vs. 96.06%), recall (98.3% vs. 71.26%), and F1 score (98.37% vs. 79.74%) using XGBoost, surpassing the results by other authors. Methodology 2 prioritizes managing missing data, using mean imputation methods integrated with “group by” for optimal grouping, to improve model performance on the 'Indian Water Quality Data' dataset by using the XGBoost. Methodology 2 performed an excellent 100% accuracy rate, which additionally solidifies the results of Ainapurea et. al. (2022), who achieved an accuracy of 98.93% by XGBoost model. Methodology 2 delivers advanced prediction capabilities for water resource management, delivering dependable resolutions for handling water quality concerns.