Optimization Analysis of Business Circle Based on Machine Learning

Abstract

This paper begins with a discussion of data preprocessing in order to improve the data quality, standardize the feature ratio, and comprehend the data pattern and trend. The preprocessing technology utilized in this paper is feature aggregation, which aggregates the transaction data based on the transaction time and date to determine the transaction type (weekday or weekend) and calculates the transaction frequency and average amount for each store as the final feature of cluster analysis. Then, for the processed data, this paper compares the advantages and disadvantages of three different clustering algorithms, including K-means clustering, spectral clustering, and DEC deep clustering, in terms of their clustering effects. The conclusion of the paper is that deep clustering is superior to traditional clustering algorithms in terms of clustering precision, but it requires more computing resources and offers practical suggestions for existing shop clustering.