The preservation of hydrobiological diversity is essential to ensuring the stability of the food chain and the sustainable development of high-Andean basins, which face increasing vulnerability to anthropogenic factors such as the construction of dams and reservoirs. In this study, multiple regression models, both linear and nonlinear, were developed to predict the Shannon–Wiener (H′) and Pielou (J′) indices of periphyton and macrobenthos using 21 water quality parameters and concentrations of nine metals in sediments. Samples of macrobenthos and periphyton were collected at seven monitoring stations during the dry and wet seasons between 2014 and 2025. For the analysis, linear regression models were compared with nonlinear machine learning models, specifically Gradient Boosting and Random Forest. Principal component analysis (PCA) revealed that variability of the basin’s ecosystem is dominated by geogenic factors (conductivity, boron, chlorides, and arsenic) and thermal influence. The Gradient Boosting model demonstrated superior predictive capacity (R² = 0.768 for macrobenthos) compared to linear models (R² = 0.354), successfully capturing the nonlinear responses of biota to stressors such as arsenic in sediments and temperature. It is concluded that natural chemical anomalies in the Titire River act as severe ecological filters, and that artificial intelligence shows promising results in the exploration of new applied tools for environmental management in extreme altoandine ecosystems.