This study analyses the performance of prediction of returns as per four Fama-French factors (market risk premium, size, value, and momentum) on six Indian portfolios (SV, SN, SG, BV, BN, BG). The study employs models based on advanced machine learning (ML) assembling techniques, namely XGBoost, LightGBM, Random Forest and Support Vector Regression (SVR) along with linear models, Ridge, and Lasso. The outputs from these models are compared with those derived from traditional OLS regression. The purpose is to evaluate the performance of factor-based models using advance machine learning algorithmic approaches. The relative contribution of each factor across models is highlighted using feature importance and SHAP value analysis. This study demonstrates that machine learning based models enhance the prediction accuracy of returns from factor portfolios, in general. However, for the large cap portfolios (BV, BN, BG), linear models like Ridge, Lasso and OLS regression perform better in explaining the variance. For the majority of models, market risk premium emerges as a consistently strong predictor. The study contributes to the existing body of literature by applying asset price theories to modern data-driven modelling in emerging markets.