Abstract: To improve the accuracy of pillar stability prediction, an improved scarab optimization algorithm（IDBO）is proposed to optimize the random forest（RF）classification prediction model. The improvement is based on5 aspects: using PWLCM chaotic mapping to enhance the quality of the initial population; introducing Levy randomwalk strategy in the rolling scarab position update to improve the global optimization ability of the algorithm; introducingan adaptive boundary control factor to weigh the global search and local development ability of the algorithm, and usinga cosine‑sine search strategy to improve the individual of the stealing scarab, enhancing its local search ability, andexecuting an out‑of‑bound handling strategy based on chaotic factors to maintain population diversity throughout theprocess; using benchmark functions for testing, which shows that the IDBO algorithm has stronger search ability androbustness; utilizing the IDBO algorithm to optimize the number of RF trees and the minimum number of leaf nodes,establishing the IDBO-RF pillar stability prediction model. Based on case analysis and comparison with existing models,the prediction accuracy of the IDBO-RF model reaches 90 %, and its prediction performance and robustness are superiorto other models.