Enhancing Surface Water Potability Assessment through a v-SVM Hybrid Statistical Learning Model

Abstract

The overall aim of the study is to improve the classification of water as either drinkable (potable) or nondrinkable. Traditional laboratory monitoring is time consuming, expensive and inefficient in monitoring quickly or in largescale. Alternatives to machine learning may be faster, although their performance is identifiable by the hyperparameter tuningprocedure, which entails the choice of model settings. This paper proposes a new hybrid algorithm, that combines v-SupportVector Regression (v-SVR) with the sparrow search algorithm (SSA) in order to achieve the task of hyperparameter tuningautomatically. The hybrid model referred to as VSRM-SSA is tested on a dataset of the water quality in the form of 3,276samples and 10 water quality items. The results depict that the VSRM-SSA model is much better than the others in terms ofhigh accuracy in classification. In the case of the data to be trained on, the accuracy was as high as 97.1 percent with G-meanand MCC being equal to 0.966 and 0.961 respectively. The model has already demonstrated good generalization capabilityusing the test data with the accuracy of 91.7% and G-mean of 0.912 and the MCC of 0.907. Obviously, these values aregreater than those that have been obtained with random search, Bayesian optimization, cross-validation, or grid search. Also,VSRM-SSA is the fastest method to compute (109 seconds) of all the methods tested. All in all, the suggested VSRM-SSAmodel offers rapid, precise, and consistent water potability classification. Its sensitivity is high and its overall performanceis equal making it promising in cases where the real-time water-quality and public-health is required.