Research on Supermarket Customer Stratification Based on Improved RFMP Model and Integrated Learning

To better predict the value of supermarket customers and classify them, this paper proposes a supermarket customer stratification model based on stacking integrated learning. Firstly, the original supermarket data set is preprocessed and an RFMP model is constructed to reflect the growth potential of different supermarket customer groups. Secondly, to eliminate the possible influence of local accuracy, the silhouette coefficient weighting method is used for correction. Finally, to give full play to the advantages of each learner and achieve the maximum prediction accuracy, the learners are pre-selected, and a stacking-based integrated learning classification model based on K-means, BIRCH, Spectral, MiniBatchKMeans and GMM is constructed, with K-means being the meta-learner. The experimental results show that, in terms of silhouette coefficient, Kalinski-Halabas index and Davidson-Bolding index scores, our model has a very good prediction accuracy improvement effect compared with the clustering effect of a single algorithm in terms of silhouette coefficient, Kalinski-Halabas index and Davidson-Bolding index scores.