An Exploratory Application of Self-Organizing Maps and Support Vector Machines for Hydrochemical Pattern Recognition and Risk Profiling in Data-Scarce Environments

Abstract

Effective water-quality assessment in data-limited environments requires analytical approaches that extract meaningful patterns from relatively small datasets. This study evaluates the complementary application of Self-Organizing Maps (SOM) and Support Vector Machines (SVM) for hydrochemical pattern recognition and exploratory risk classification using physicochemical and heavy-metal data from water sources in Idemili South and Ekwusigo Local Government Areas of Anambra State, southeastern Nigeria. Compared with the World Health Organization (WHO) and Nigerian Standard for Drinking Water Quality (NSDWQ) guidelines, the water-analysis results revealed significant heavy-metal contamination by Hg, Cd, and Pb. The maximum concentrations of Hg and Cd exceeded the applicable guideline values by approximately 94-fold and 36.3-fold, respectively. SOM analysis identified seven distinct hydrochemical clusters, with Hg and Cd emerging as the most influential contaminants in the observed water-quality patterns. Spatial analysis of the composite water-quality indices further identified localized areas of elevated contamination, particularly in the northwestern and southeastern parts of the study area. The SVM categorized seven samples as Low Risk and four as Medium Risk, achieving a classification accuracy of 81.82%. Integrating SOM for unsupervised pattern recognition and SVM for exploratory supervised classification provided insights into hydrochemical variability and risk patterns within the limited dataset. Although limited by a small sample size, the study demonstrates that combining water-quality indices with machine learning provides a viable framework for exploratory contamination studies and water-resource evaluation in data-limited settings.