Abstract: With the increase of coal mining depth, rock burst occurs frequently, and accurate prediction of rock burst hazard level is very important for coal mine safety mining. However, there are many indicators affecting rock burst, and the data characteristics are different. Aiming at the problem that a single algorithm cannot meet the requirements of different indicators, a prediction method of index information fusion model based on data mining was proposed. The model consists of conditional K nearest neighbor, a Bonus random forest and a weighted voting algorithm. Firstly, the buried depth index is input into the conditional K nearest neighbor algorithm. Then, other indicators were input into the random forest algorithm, and the two algorithms were used to predict the hazard level of rock burst. Finally, the prediction results of these two algorithms are put into the weighted voting algorithm for voting, and the risk level with the highest weighted result is the final danger level. In this paper, data mining and model construction were carried out using 350 samples collected from 10 working faces in Qianqiu Coal Mine from 2010 to 2013. The experimental verification was carried out using 20 sample data collected from 4 working faces in Qianqiu Coal Mine from 2015 to 2016. The results show that the prediction method in this paper has higher prediction accuracy than the traditional prediction method, which provides a certain reference for the early warning of rock burst hazard level in coal mines.