To ensure an effective management of circulation loss, a critical challenge within drilling operations, four balanced datasets was developed aimed at categorizing varying severity levels of circulation loss at Iran's Marun Oil Field. Employing diverse cross-validation strategies and setting multiple average performance metrics as the optimization objectives, optimal hyperparameters was identified across several models, subsequently determining the best model and its parameter setup. It was found that the selected optimal model exhibited significant enhancements in accuracy, recall, and F₁-scores. Specifically, different models, such as extreme gradient boosting and categorical gradient boosting, were applied to datasets corresponding to various well leakage degrees, with all performance metrics surpassing the 0.95 threshold, thereby illustrating the models' superior performance. Moreover, by computing shapley Additive explanations values, the study provided deeper insights into the predictive behaviors, underscoring the substantial influence of hyperparameter optimization on model performance enhancement. Additionally, model interpretation emerged as crucial for a profound understanding of the predictive mechanisms, highlighting the necessity of employing such techniques in contemporary research to elucidate model functionalities and outcomes.