This study examined the effect of data preprocessing on the performance of machine learning models in forecasting monthly precipitation in the Lake Namak basin. CHIRPS data were first validated against ground based observations (NSE = 70%), then cleaned using the Z Score method to remove outliers and subsequently reconstructed using Singular Spectrum Analysis (SSA) and linear interpolation. Statistical analyses indicated that preprocessing substantially reduced the skewness and kurtosis of the time series and enhanced data stability without noticeably altering the variance. Performance comparisons among the Random Forest, XGBoost, SVR, and CatBoost models showed that preprocessing significantly improved forecasting accuracy; notably, the Random Forest model achieved the largest enhancement, with MSE decreasing from 28.7 in the original series to approximately 15.7–16.0 in the corrected series, while SVR exhibited the least sensitivity to data adjustments. Both visual and numerical evaluations further confirmed the superior performance of Random Forest and XGBoost in terms of higher correlation and lower bias. Overall, the findings demonstrate that integrating effective preprocessing techniques with machine learning models—particularly Random Forest—plays a critical role in improving precipitation forecasting accuracy and reducing hydrological uncertainties.