Reviewing artisanal small-scale gold mining driven heavy metal pollution in Ghana through principal component analysis and partial least squares regression: A machine-learning approach

Abstract

This study applied multivariate and machine-learning approaches to synthesise evidence on mercury, arsenic, cadmium and lead contamination from Artisanal and Small-Scale Gold (ASGM) mining across Ghana. A total of 21 studies and 57 medium combinations met the eligibility criteria and were included. A review of mean concentrations of the Potentially Toxic Elements (PTEs) was carried out across studies from both Northern and Southern Ghana. Water showed the lowest concentrations, yet 78% of samples still exceeded World Health Organization (WHO) and Ghana Standards Authority (GSA) limits. Sediment had the highest levels, with 91% classified as heavily contaminated. In soils, 63% fell within moderate to high contamination classes, also above recommended thresholds. Geochemical controls explain the contrasts across media. These concentrations are likely influenced by shifts in soil and sediment chemistry including pH, redox conditions, soil texture, mineralogy and rainfall. Principal Component Analysis (PCA) accounted for 42.2% of the total variance. PC1 scores were then modelled using Support Vector Regression (SVR) with region, media type, and publication year as predictors. The SVR model achieved an R² of 0.233 and a Root Mean Square Error (RMSE) of 0.93, explaining about 23.3% of the variance in PC1. For validation, Partial Least Squares Regression (PLSR) was applied using Region, Media and Year as predictors. Its first latent component was significant (p < 0.001) and explained 24.8% of the variance in PC1. The findings support SDG 6.1, underscoring the need for stronger water-safety measures. They also advance SDGs 6.3, 12 and 15 by highlighting contamination sources, geochemical drivers and operational gaps that can guide pollution control, sustainable mining and targeted environmental management.