Using Unsupervised Machine Learning Methods for Analyzing Customer Habits

Osnovni zadatci nenadziranog strojnog učenja su grupiranje, procjena gustoće, otkrivanje anomalija i stršećih vrijednosti i smanjenje dimenzionalnosti. Nenadzirano strojno učenje ima veliku primjenu u marketingu. Danas to područje obiluje podatcima pomoću kojih tvrtke mogu razumjeti potrebe kupaca i steći njihovu naklonost, a time i dugoročni profit. U ovom je radu naglasak na metodama grupiranja kupaca na temelju njihovih karakteristika i potrošačkih navika. Na odabranom podatkovnom skupu provedena je eksploratorna analiza, reducirana dimenzionalnost te su primijenjeni algoritmi K-sredina, K-medoida i algoritam hijerarhijskog grupiranja (HAC). Uočeno je da na ovom podatkovnom skupu algoritam K-sredina dao bolje rezultate od algoritma K-medoida: dobivene 3 grupe su bile homogenije. Na temelju razlika između grupa kupaca tvrtka bi mogla personalizirati buduću marketinšku kampanju. Unsupervised machine learning models are utilized for four main tasks: clustering, density estimation, anomalies and outliers detection, and dimensionality reduction. Nowadays, there's a lot of data that can be used to satisfy customer needs and sustain a profitable business. Therefore, unsupervised machine learning plays a major role in marketing. The main focus of this paper is customer clustering techniques. Before performing customer segmentation based on their characteristics and purchase behavior, an exploratory analysis of chosen dataset has been done. Clustering algorithms that were used in this paper are the K-means algorithm, the K-medoids algorithm, and hierarchical agglomerative clustering (HAC). It turned out that the K-means algorithm outperformed the K-medoids algorithm on this dataset: 3 obtained clusters were more homogeneous. Based on differences among clusters company could personalize its future marketing campaign.