Protein plays an important role in the cellular process of an organism, and its function is demonstrated by protein interaction. Rich information on protein interactions can facilitate the treatment of diseases and the development of drugs, so accurate prediction of protein interactions is of great significance. High‐flux biological experiments can be used to predict new protein pairs, but they are expensive and time‐consuming to operate and do not meet the demand for such information. With the rise of machine learning algorithms and the increasingly powerful computing power, the use of scientific computing models to predict each other has become the first choice. This paper mainly studies the application of weighted sparse representation classifiers under protein sequence feature coding. First of all, the composition, transfer and distribution of the physical and chemical properties of amino acids are selected to encode the amino acid sequence. Secondly, according to the characteristic importance of random forest, the feature operator de‐dimensionally de‐noises. Finally, for the features extracted in this paper, a weighted sparse representation classifier with strong noise resistance is used to classify the feature set. The results of the 50% cross‐validation were: accuracy 96.97%, sensitivity 97.51%, accuracy 96.43%, Matthews correlation coefficient 93.91%, Predictive results are better than existing machine learning models.