This article, which results from a sophisticated pavement condition assessment process, conducts a comparative analysis of machine learning models, focusing on stack regressor model which combines CatBoost and XGBoost as base estimators and linear regression as the final estimator. This method allows for exceptionally challenging data manipulation and feature interactions due to the unique properties of CatBoost and XGBoost utilizing key inputs such as cracking percentage, plasticity index, maximum dry density, California bearing ratio (CBR), soil type, and layer thickness. A dataset of 2001 samples taken from the Indian road network is trained and validated. The result clearly shows that the stacking regressor model outshines the classical methods and other machine learning algorithms, such as random forest, CatBoost, XGBoost, and LightGBM, out with the highest R2 scores (0.8055 for SCI and 0.7753 for BCI) and comfort with the lowest MSE (0.0208). The k-fold validation MSE for stacking regressor model was found to be 0.0208 which shows that the stacking regressor model generalizes best for unseen data than standalone models, with CatBoost and XGBoost achieving slightly lower performance. SHAP analysis was performed which showed cracking (%) being the most impactful feature among others and its high values push model output positively. This work presents the possibilities of combining advanced methods of machine learning in road management, which guarantee higher precision and speed, as well as cost-saving compared to traditional methods, i.e., the Flexible Weight Deflectometer (FWD) testing.