Aplicação da clusterização por K-means para criação de sistema de recomendação de produtos baseado em perfis de compra

Abstract

The use of predictive machine learning models for big data is today one of the main trends to be explored by data science. Its application to the business world for a search of competitive differential is directly related to Business Intelligence so companies can make more assertive decisions. Thus, this paper proposes to apply a machine learning technique to create a product recommendation system based on customers' purchase profile, modeled for a product distribution company. For this purpose, the K-means clustering algorithm was used to group customers based on their purchase profile. Finally, the recommendation system's principle is based on a comparative analysis between customers in the same cluster and based on their geographic distances to recommend that item that sells well in one point of sales but does not perform so well in another. At the end of the application 70 clusters were generated for the entire range of customers of the company focused in the present study. Each customer in each cluster received a list containing 5 recommended products based on the comparison made with their close neighbors of similar buying profile.