India's extensive road network plays a crucial role in national development, yet faces persistent challenges due to inadequate maintenance, particularly in rural and urban segments. This study integrates calibrated HDM-4 deterioration models with machine learning (ML) techniques to improve the prediction and management of pavement distresses such as cracking and potholes. Data were collected from 20 road sections spanning 108 km across Madhya Pradesh, with performance indicators analyzed over a five-year period. Calibration of HDM-4 parameters for both rural and urban sections revealed discrepancies between model predictions and field conditions, especially under varying environmental and traffic stressors. To enhance predictive capability, machine learning models—Random Forest, Gradient Boosting, and Artificial Neural Networks—were applied. The Random Forest model demonstrated the highest accuracy for cracking prediction (R² = 0.95), while ANN effectively captured pothole progression with an R² of 0.89. The Box-Cox transformation and statistical assumptions were addressed to ensure model robustness. The findings underscore the importance of combining empirical and data-driven approaches for reliable pavement management, enabling better maintenance scheduling and resource optimization.