Rockburst data category imbalance often leads to a decline in the prediction performance of intelligent rockburst prediction models. To address this issue, this study employs Adaptive Synthetic Sampling (ADASYN) to balance the collected dataset of 269 rockburst sets. An intelligent prediction model is constructed by integrating the whale optimization algorithm (WOA) with the light gradient boosting machine (LightGBM) model. This model uses six feature indexes as input variables: surrounding rock pressure ( \({\sigma }_{\uptheta }\) ), uniaxial compressive strength ( \({\sigma }_{\text{c}}\) ), uniaxial tensile strength ( \({\sigma }_{\text{t}}\) ), elastic energy index ( \({w}_{et}\) ), brittleness coefficient ( \({\sigma }_{\text{c}}/{\sigma }_{\text{t}}\) ) and stress coefficient ( \({\sigma }_{\uptheta }/{\sigma }_{\text{c}}\) ). The model's performance was systematically evaluated using five-fold cross-validation and multiple evaluation metrics, with comparisons made against various optimization algorithms and comparative models. The experimental results indicate that the ADASYN-balanced WOA-LightGBM model achieved an average prediction accuracy of 0.9071, demonstrating excellent discriminative performance. Further validation using 13 case sets from practical engineering projects such as the Jiangbian Hydropower Station preliminarily confirms the model's good potential for engineering application. The method proposed in this study provides an effective novel approach for data-balance-based intelligent risk assessment of rockburst.
