The lost circulation incident in drilling operation has the characteristics of sudden occurrence and difficulty in treatment， which leads to low success rate of plugging and high cost. In order to solve this kind of problem quickly， accurately and efficiently， we must accurately predict lost circulation and put forward plugging measures. In this paper， the gradient boosted decision tree algorithm （GBDT） is used to study the loss circulation prediction model with the feasibility of GBDT as a lost circulation prediction model verified by analyzing some machine learning algorithms. The lost circulation case decision base is established by using case-based reasoning technology， and the Euclidean algorithm and LB_Keogh algorithm are used to search lost circulation cases to put forward corresponding lost circulation treatment measures. The reliability of similarity calculation has been verified through field cases.