<p>This study focused on the Daluxi River, a small watershed and a primary tributary of the Yangtze River. Based on the nonlinear characteristics of water quality parameters and environmental factors such as meteorological and hydrological influences, a comparative analysis was conducted using Kernel Principal Component Analysis (KPCA) and Principal Component Analysis (PCA). KPCA extracted four potential sources for both the upstream and downstream sections, accounting for 79% of the total variance in each case—an increase of 7% and 6% compared to PCA, respectively. To address the limitation of KPCA in directly revealing the relationship between principal components and the original water quality data, six machine learning algorithms—Extreme Learning Machine (ELM), Backpropagation Neural Network (BPNN), Support Vector Regression (SVR), Decision Tree (DT), Random Forest (RF), and Gradient Boosting Decision Tree (GBDT)—were employed to perform regression analysis between the kernel principal components and the original water quality parameters, thereby elucidating source characteristics. The results indicated that GBDT exhibited the best fitting performance (R² = 0.988, MAE = 0.05, RMSE = 7.13%). Based on the extracted KPC, the Absolute Principal Component Score-Multiple Linear Regression (APCS-MLR) model was used to calculate the contribution rates of various pollution sources in the Wandang and Siming areas. The results indicate that combining KPCA with GBDT and APCS-MLR can effectively uncover the complex relationships among water quality, meteorological, and hydrological factors, thereby enhancing the accuracy and reliability of pollution source analysis. This study advances research by using KPCA to capture nonlinear relationships and integrating machine learning for enhanced pollution source analysis.</p>