HYBRID ONE-DIMENSIONAL CNN–SVM AND ATTENTION-BASED BI-GRU MODELS FOR DRINKING-WATER POTABILITY CLASSIFICATION WITH MISSING-DATA IMPUTATION

The present study developed two missing-data-aware deep-learning pipelines for drinking-water potability classification and compared them with deep and conventional baselines on a public dataset of 3,276 samples and nine physicochemical parameters. The main conclusions are as follows: (1)     Missing values affected three parameters (pH 14.99%, sulphate 23.84% and trihalomethanes 4.95%; 4.86% of all parameter values). Interpolation-based MVM and KNN imputation allowed the complete dataset to be used for training and evaluation. (2)     The hybrid 1D-CNN–SVM achieved the highest test-set performance (accuracy 0.999, precision 0.997, recall 0.998 and F1-score 0.999), outperforming logistic regression by 0.229 in accuracy, while the random forest collapsed largely towards majority-class prediction (accuracy 0.63, precision 0.39). (3)     The ranking MLP < LSTM < GRU < Bi-GRU < CNN–SVM was preserved for all sample sizes. The CNN–SVM reached 84% accuracy with 100 samples and 92% with 500 samples, and a recall of 99.56% with 800 samples; the attention-based Bi-GRU followed closely (82%, 91.2% and 99.34%). (4)     Accuracy increased with the number of input parameters for all models, while the proposed models were the least sensitive to reduced inputs (CNN–SVM: 87% with one parameter, recall 98% with all parameters), which favours deployment with low-cost sensor subsets. (5)     The RBF kernel yielded the highest testing correlation coefficient (0.801) among four SVM kernels, although its larger training–testing gap in MSE indicates scope for stronger regularisation.

[1]      K. Kalaivanan and J. Vellingiri, "Survival study on different water quality prediction methods using machine learning," Nature Environment and Pollution Technology, 2022. [2]      N. Najwa Mohd Rizal, G. Hayder, M. Mnzool, B. M. Elnaim, A. O. Mohammed, and M. M. Khayyat, "Comparison between regression models, support vector machine (SVM), and artificial neural network (ANN) in river water quality prediction," Processes, 2022. [3]      A. Kuthe, C. Bhake, V. Bhoyar, A. Yenurkar, V. Khandekar, and K. Gawale, "Water quality prediction using machine learning," International Journal of Computer Science and Mobile Computing, 2023. [4]      Y. Yang, Q. Xiong, C. Wu, Q. Zou, Y. Yu, H. Yi, and M. Gao, "A study on water quality prediction by a hybrid CNN-LSTM model with attention mechanism," Environmental Science and Pollution Research, vol. 28, pp. 55129–55139, 2021. [5]      Q. Zhang, R. Wang, Y. Qi, and F. Wen, "A watershed water quality prediction model based on attention mechanism and Bi-LSTM," Environmental Science and Pollution Research International, vol. 29, pp. 75664–75680, 2022. [6]      L. Zhang, J. Xie, X. Liu, W. Zhang, and P. Geng, "Research on water quality prediction based on PE-CNN-GRU hybrid model," E3S Web of Conferences, 2023. [7]      X. Wu, Q. Zhang, F. Wen, and Y. Qi, "A water quality prediction model based on multi-task deep learning: A case study of the Yellow River, China," Water, 2022. [8]      M. O. Khairandish, M. Sharma, V. Jain, J. M. Chatterjee, and N. Z. Jhanjhi, "A hybrid CNN-SVM threshold segmentation approach for tumor detection and classification of MRI brain images," IRBM, vol. 43, no. 4, pp. 290–299, 2022. [9]      D. J. Chaudhari and K. Malathi, "Detection and prediction of rice leaf disease using a hybrid CNN-SVM model," Optical Memory and Neural Networks, vol. 32, pp. 39–57, 2023. [10]    H. A. Madni, M. Umer, A. Ishaq, N. Abuzinadah, O. Saidani, S. Alsubai, M. Hamdi, and I. Ashraf, "Water-quality prediction based on H2O AutoML and explainable AI techniques," Water, 2023.