<p>Faulting is one of the critical structural failures in jointed plain concrete pavements. The repercussions of this defect are an increase in maintenance cost, an impact on safety and a decrease in the overall performance of the pavement. A Machine Learning (ML) based algorithm i.e. Extreme Gradient Boosting (XGBoost) was utilised to develop a faulting prediction model in Jointed Plain Concrete Pavement (JPCP). This model was also optimized by using suitable optimizing functions, such as randomized grid search and Bayesian optimization to tune the hyper parameters. The best-performing model, in combination with a suitable optimizer, was determined in this study. It was observed that Extreme Gradient Boosting, tuned using Bayesian Optimization, provided the best prediction results for the validation dataset. Furthermore, a sensitivity analysis was carried out utilizing partial dependence plots to gain insights into the behavior of various features. The relative variable importance of the features was also determined for this model to identify the key variables. Joint spacing, tensile strength and slab thickness were found to be most influencing parameters. Overall, the findings not only provide valuable insights into faulting prediction and variable importance but also lay the foundation for novel approaches to address the critical issue of faulting in jointed plain concrete pavements.</p>