An optimized explainable artificial intelligence approach for sustainable clean water

Abstract

Water, sanitation, and hygiene are essential components of the 2030 agenda for sustainable development. Goal 6 is dedicated to guarantee all societies have access to water and sanitation. Water quality (WQ) assessment is crucial to ensure the availability of clean water. This paper presents an approach called AHA–XDNN for predicting WQ. The proposed approach is based on three pillars to predict WQ with high accuracy and confidence, namely, deep neural networks (DNN), artificial hummingbird algorithm (AHA), and explainable artificial intelligence. The proposed approach involves five phases: data preprocessing, optimization, training, and evaluation. In the first phase, problems such as unwanted noise and imbalance are addressed. In the second phase, AHA is applied to optimize the DNN model’s hyper-parameters. In the third phase, the DNN model is trained on the dataset processed in the first phase. The performance of the optimized DNN model is evaluated using four measurements, and the results are explained and interpreted using SHapley additive exPlanations. The proposed approach achieved an accuracy, average precision, average recall, average F1-score of 91%, 91%, 91.5%, and 91% on the test set, respectively. By comparing the proposed approach with existing models based on artificial neural network (ANN), the proposed approach was able to outperform its counterparts in terms of average recall and average F1-score.