A Comparative Study of Deep Learning Models for Water Quality Prediction

Doaa El-Shahat, Department of Computer Science, Faculty of Computers and Informatics, Zagazig University, EgyptFollow
Mohamed Ismail, Department of Computer Science, Faculty of Computers and Informatics, Zagazig University, EgyptFollow

ORCID

Doaa El-Shahat: https://orcid.org/0000-0003-1681-5039

Mohamed Ismail: https://orcid.org/0009-0008-8956-254X

Keywords

Water quality prediction, Deep learning, Feature extraction, Convolutional neural network, Informer encoder

Article Type

Original Article

Abstract

One of the most important challenges in the recent period, and one that still presents a significant challenge, is sustainable management of water resources and accurate prediction of water quality to address increasing pollution and climate changes. Therefore, our study aims to address the complexity and noise in water quality data with time dependencies in water quality data in order to provide a more accurate prediction of water quality by evaluating and comparing six of the latest and most powerful deep learning models that have been recently developed, namely: VBAED, SA-LSTM-LOADEST, LWQformer, ESWT-IE-SBiLSTM, DPSGT, in addition to the CNN-CRNN-M5T model. We tested these models on the Global Water Quality Science dataset, which contains 2.82 million measurements for eight water quality parameters. The models were evaluated based on several performance metrics, including accuracy, balanced accuracy, recall, F1 score, ROC-AUC scores, precision, and confusion matrix. These metrics demonstrated the high efficiency of all six selected models. However, the LWQformer model outperformed all others, achieving 98.56% accuracy and 97.78% balanced accuracy. It was closely followed by the SA-LSTM-LOADEST model, which achieved 98.53% accuracy and 97.69% balanced accuracy. These percentages reflect the close performance of these two models, highlighting the strength of attention-based models and transformers due to their ability to handle temporal changes and their high capacity for extracting complex features with extreme accuracy enables us to accurately and early predict water quality and protect the aquatic environment.

How to Cite

El-Shahat, Doaa and Ismail, Mohamed (2026) "A Comparative Study of Deep Learning Models for Water Quality Prediction," Sustainable Machine Intelligence Journal: Vol. 14: Iss. 2, Article 1.
DOI: https://doi.org/10.63689/3005-3617.1086