Abstract: To solve the problems of time-consuming and costly wet experiments for drug development, machine learning has been applied to the prediction of drug-target interactions. At the same time, in order to solve the constraints of machine learning in building drug-target interaction models, it is subject to various problems such as class imbalance of classifiers and parameter optimization. The paper proposes a drug-target interaction prediction method (SEELM-DTI) based on a spherical evolution extreme learning machine, which mainly uses a screening method to select high confidence negative samples and a spherical evolution to optimize the parameters of the extreme learning machine. The researcn compared SEELM-DTI with SELF-BLM, NetLapRLS, WNN-GIP, SPLCMF, BLM-NII in a benchmark dataset and evaluated the metrics of AUC and AUPR.The experimental results showed that the SEELM-DTI outperformed other benchmark algorithms and solved the class imbalance and parameter optimization. Finally, the effectiveness of SEELM-DTI in predicting drug-target interactions was validated on several commonly used drug databases.