Integrating multivariate statistics and AI for groundwater heavy metal risk detection: a case study from Yilan, Taiwan

ABSTRACT Heavy metal contamination in groundwater continues to threaten water security and public health, particularly in regions with complex hydrogeology such as Yilan County, Taiwan. Conventional monitoring approaches, often restricted to single statistical methods or regulatory thresholds, are insufficient to uncover potential hazards or link them to health risks. To address these limitations, this study applies an integrated, risk-oriented analytical approach that first employs multivariate statistical analysis (PCA/FA) to screen and interpret representative groundwater quality parameters, and then integrates artificial intelligence models, applying isolation forest for anomaly detection, followed by random forest for supportive verification and parameter importance assessment. This stepwise integration supports interpretability and consistent risk screening in routine groundwater monitoring. Cross-method synthesis consistently identified chromium (Cr), arsenic (As), and cadmium (Cd) as core contaminants, validated by health risk indices (HQ and CR), while other metals (Ba, Cu, Pb, Se, In, Hg, and Sb) were considered secondary indicators. The integrated analysis reduced false-negative risk during anomaly screening and supported early identification of samples requiring further attention. By embedding HQ/CR indices into the workflow, the approach moves beyond compliance-based assessment and provides a transferable decision-support tool for groundwater monitoring, early-warning support, and practical water management.