Early warning of bond default risk is to predict the future bond default status according to the enterprise's financial factors, non-financial factors and external macro factors. For different combinations of variables, the effect of default prediction is different, and there is bound to be an optimal combination of indicators, which can minimize the error of default prediction。For different critical points of default judgment, the effect of default prediction is different, and there is bound to be an optimal critical point of default judgment, which can distinguish the bonds that default or not to the greatest extent. This paper uses the random forest model to select the feature combination, and studies the default risk of bonds based on Logit model. The first contribution of this paper is that in the selection of the optimal feature combination, for different decision tree numbers, under the premise of the minimum error of the second kind, an optimal random forest is obtained by AUC maximum backstepping; Through the comparison of different combinations of node feature in the optimal random forest, the feature combination with the largest AUC is found. Second, in the determination of the optimal default judgment critical point, for the different proportion of the first type of error and the second type of error, the minimum weighted sum of the first type of error and the second type of error is taken as the objective function to deduce the optimal critical point of logical regression.Third, the prediction accuracy of this model is higher than that of seven popular big data prediction models, such as support vector machine model, gradient boosting iterative decision tree, and neural network. Based on the data of bonds issued by Chinese bonds listed companies during 2014-2018, the empirical research shows that the key features that have an impact on China's medium and short-term default prediction are: "monetary capital / short-term debt", "net profit", "the number of bonds issued by issuers", "industry prosperity index" and "industry entrepreneur confidence index". The key features that have an impact on short-term default prediction are: "monetary assets", "quick ratio", "fixed asset investment price index" and "money supply M0". The key features that have an impact on the medium-term default forecast are "registered capital of the issuer", "repayment amount of bonds at maturity" and "bond maturity index".