This paper uses a clustering algorithm to extract and classify credit rating features of tobacco retailers and evaluates whether the classification results are reasonable by combining clustering evaluation indexes. The distance between samples is calculated using the similarity measure. The natural domain method density and peak clustering method are used to analyze the distribution of sample points in the data set. Combining the cluster analysis creates the tobacco retail credit rating evaluation index. The results show that cluster analysis can effectively extract credit rating features from tobacco retailers. When the number of features is 25, the model has the best classification effect, with a classification accuracy rate of 91.1%, a recall rate of 91.5%, and an F1 value of 91.3%. The classification of tobacco retailers’ credit ratings can be improved effectively by the research in this paper.