[Objective]In order to ensure the objectivity, timeliness and accuracy of the overall budget performance evaluation of research institutions, and improve the efficiency of performance evaluation.

[Methods]This paper proposes a LightGBM-based overall budget performance evaluation and prediction method, and integrates multi-source data of scientific research management information system. Based on the correlation between research input/output data and research performance, machine learning is used to analyze and predict the overall budget performance evaluation results of scientific research institutions.

[Results]In the the application of the overall budget performance evaluation of scientific research institutions, the accuracy of the performance evaluation prediction method proposed in this paper is 91.94%, the human resources required for the process of budget performance evaluation is reduced from 10 people to 5 people, and the time cost is reduced from 1 month to 2 months to about 10 days.

[Limitations]Some performance evaluation indicators do not have corresponding data items in the scientific research management information system.

[Conclusions] The proposed method has excellent performance in the prediction of the overall budget performance evaluation results. It can reduce the fairness problems caused by subjective evaluation, save the human resources and time cost of budget performance evaluation, and improve the efficiency of performance evaluation.