Groundwater Quality Prediction Using Deep Neural Networks with Principal Component Analysis Based Feature Reduction and K-Fold Cross Validation

Groundwater quality plays a vital role in meeting domestic, agricultural, and industrial needs. However, the intensification of human activities has led to the degradation of groundwater quality, necessitating the development of accurate and efficient predictive methods. This study aims to develop a groundwater quality prediction model using a DNN optimized with PCA and K-Fold Cross Validation. The dataset comprises various physical and chemical parameters of groundwater collected from Yogyakarta City. Dimensionality reduction through PCA was employed to address data complexity and mitigate the risk of overfitting, while K-Fold Cross Validation was implemented to enhance the model's reliability and generalization capabilities. The findings show that the baseline DNN model attained an accuracy of just 46.63 %. However, after incorporating PCA and K-Fold Cross Validation, its performance rose sharply, reaching an accuracy of 97.67 %. Moreover, precision, recall, and F1-score values were nearly perfect for all groundwater quality classifications, namely compliant, lightly polluted, moderately polluted, and heavily polluted. The best model performance was achieved using Optimizer 2 Model 2 at Fold 3, demonstrating an accuracy of 97.67 % and stable prediction across categories. The study's outcomes indicate that the DNN-PCA-KFold framework offers strong potential for real-time groundwater quality monitoring, enabling early contamination detection and fostering adaptive, data-informed water resource management. Furthermore, the findings pave the way for wider adoption of AI-driven predictive models within the environmental domain.