Segmentation and prediction of customer behavior based on the improved RFM model (LRFMSP)

In recent years, with the development of machine learning and big data technology, user data has become an important element in the production processes of companies. By applying data mining approaches to customer data, organizations understand customers' behavioral patterns, their needs, and the hidden relationships in the data, and based on these patterns, they can better use their resources to meet customer needs. Clustering is one of the data mining techniques used to group customers according to their different characteristics. The main goal of this research is to cluster customers based on LRFMSP indicators and finally classify them and predict their buying behavior using decision tree (DTC), multilayer perceptron (MLP) and support vector machine (SVM) classification techniques. The study was conducted on 387,496 transactions from customers of a retail store in Western Europe between February 2018 and February 2019. Each transaction attributed to a customer is part of an individual's behavior that is modeled on a set of transactions to shape the customer's purchasing behavior. Performing K-means++ clustering and determining the optimal K led to identifying three clusters for customers. Also, testing and checking the classifiers showed that the MLP classifier with one hidden layer and six neurons in this layer would be the most accurate and the DTC classifier is the fastest among the classifiers reviewed. Examining the behavior of cluster customers showed that customers can be divided into three categories: loyal, potential, and lost.