Integrating principal component analysis and machine learning to assess riverine impacts on lake water quality: a case study of the Bilate River–Lake Abaya Watershed, Rift Valley, Ethiopia

ABSTRACT Rivers are dynamic systems that transport nutrients, sediments, and pollutants to downstream lakes, contributing to water quality deterioration and ecosystem stress. This study aimed to assess seasonal and spatial variations in water quality within the Bilate River (BR)–Lake Abaya (LA) system and quantify the river's influence on the lake. Principal component analysis (PCA) was applied to identify key drivers of water quality, while after comparison, RF and XGBoost models were used to predict target parameters. SHAP was employed to interpret model outputs by quantifying feature contributions. Statistical analyses revealed significant differences across both seasons and transects (p < 0.05). PCA showed that in the dry season, PC1 explained 51.3% of variability, driven by TDS (0.85), DO (0.75), NO3− (0.65), and PO4−3 (−0.89). In the wet season, PC1 explained 40.5% of variability, with pH (0.87), TDS (0.78), salinity (0.74), TSS (−0.83), and BOD5 (−0.78). Model evaluation showed that XGBoost achieved higher predictive accuracy for DO (CA = 0.19,0.30), NO3− (CA = 0.22,0.25), and PO4−3 (CA = 0.13,0.15), while RF performed better for TSS (CA = 0.08,0.13). SHAP analysis showed that LA's water quality, particularly TSS, DO, NO3−, and PO4−3, is strongly influenced by BR inputs. These impacts are linked to intensive agriculture and land-use changes. Therefore, restoration of wetlands and implementation of soil conservation strategies are recommended to preserve LA's water quality.