Cluster-Based Federated Learning for Automotive Customer Profile Analysis

Abstract

In the context of increasingly fierce competition in the automotive market, car dealerships are shifting from a "product-oriented" approach to a "customer-oriented" one. This paper establishes a customer profile tagging system based on car purchasing behaviors and customer attributes, and constructs a customer profile model through cluster analysis. However, traditional machine learning algorithms face limitations such as insufficient data, single sources, and incomplete tags. Additionally, data cannot be directly shared due to data security concerns. Therefore, this paper applies a FederatedKMeans algorithm, which combines federated learning and K-means clustering, to construct profiles. The superiority of this algorithm in terms of performance is demonstrated through empirical tests, and it is used for cluster analysis to categorize customers into different groups, providing corresponding marketing strategies.