Oxidative stress can lead to various diseases, so achieving antioxidation is crucial for combating these diseases. However, the study of antioxidant mechanism is still in its early stage. Most current methods measure the antioxidant properties of polypeptides using exptl. approaches, which are typically time-consuming and laborious. It is therefore urgent to develop a theor. method to identify antioxidant activity based on peptide sequence information. In this study, amino acid composition, dipeptide composition, and "one-hot" vectors were used to characterize the sequence information of tripeptides. Machine learning methods such as support vector machine, random forest, gradient tree boosting and multi-layer perceptron were utilized to construct models for identifying their antioxidant activities. For the five collected datasets, the Q2 of the random forest model reached 0.864, 0.903, 0.867, 0.611, and 0.703, resp. Compared with the existing methods 0.061, 0.105, 0.028, 0.121 and 0.111 improvements have been obtained, demonstrating the effectiveness of the current approach. The developed method is expected to significantly contribute to the study of polypeptide antioxidant mechanisms.