Abstract: Groundwater fluctuations in colluvial deposits induced by rainfall infiltration play a critical role in slope stability. Developing reliable models for groundwater response to rainfall infiltration enhances understanding of the mechanisms underlying rainfall-induced shallow landslides in bedrock mountain areas. This study utilized rainfall and groundwater monitoring data collected in Qingchuan County, Sichuan Province, from 2019 to 2023 to systematically analyze the response patterns of slope groundwater levels during rainfall infiltration. A support vector regression (SVR) model was developed to predict the saturation depth ratio of slopes. To address the systematic lag and peak underestimation inherent in conventional SVR models during infiltration response prediction, this study introduced lag features, differencing features, rolling statistical features, periodic feature optimization, and extreme value feature optimization, constructing a multi-dimensional feature-optimized SVR model. Results demonstrate that the optimized SVR model achieves substantially improved predictive accuracy, with the coefficient of determination (R²) on the test set increasing from 0.681 to 0.880, while root mean square error and mean absolute error decrease to 0.0405 and 0.0270, respectively. The proposed optimization model more effectively captures the nonlinear groundwater response processes in colluvial slopes during extreme rainfall events, demonstrating its potential as a practical tool for rainfall-induced landslide assessment and early warning.