The accurate and reliable simulation and prediction of runoff in the Beas River Basin are becoming more and more important due to the increased uncertainty posed by climate change, which is making it difficult to manage water resources efficiently. In order to minimize the effect of flash floods, estimating the accurate peak flow is essential. It can be challenging to comprehend and anticipate peak flow due to natural streamflow variance as well as the streamflow management offered by dams and reservoirs. Which makes it difficult to mimic hydrologic behavior on a daily scale with reliable accuracy. This study evaluated the efficacy of physics-aided machine learning (ML) based regression models for modeling streamflow in combination with process-based hydrological SWAT (soil and water assessment tool). Performance of eight machine learning (ML) models including linear regression (LR), multi-layer perceptron (MLP), light gradient-boosting machine (LGBM), extreme gradient boosting (XGBoost), kernel ridge (KR), elastic net (EN), Bayesian ridge (BR), and gradient boosting (GB) have been analyzed and compared with the calibrated-SWAT (cSWAT) model. The Nash–Sutcliffe efficiency (NSE), root mean square error (RMSE) and coefficient of determination (R²) were used to assess the effectiveness of both models. Results showed that the uncalibrated SWAT in combination with ML regression models (cSWAT-ML) performed well and found comparable to calibrated SWAT (cSWAT), though, few ML regression models in combination with uncalibrated SWAT (uSWAT-MLmodels models) performed superior. cSWAT model performed well with R² values of 0.73, RMSE value of 276.92 m³/s and NSE value of 0.72. In uSWAT-ML, EN and BR have obtained better results with R² values of 0.89 and 0.89, NSE values of 0.87 and 0.87, and RMSE values of 158.31 m³/s and 159.48 m³/s. The approached uSWAT-ML models have effectively predicted the peak stream flow rates with models BR and EN have predicted with better results of R² value of 0.71 each. This study’s findings highlight the potential of all the eight ML models as promising techniques for predicting the peak flow discharge values when uncalibrated process-based models are combined.