Predviđanje prodaje na platformi e-trgovine, korišćenjem data mining modela

Abstract

In this paper we applied twinning algorithm for product that are sold via e-commerce platform. To establish relatively homogenous product groups that were on sale on this e-commerce platform during the last year, it was necessary to form predictive mathematical model. We determined set of relevant variables that will represent group attributes, and we applied K-means algorithm, Market Basket model and Vector Distance model. Based on analysis of basic and derived variables, fixed number of clusters was introduced. Silhouette index was used for the purposes of detecting whether these clusters are compact. Using these cluster separations, we created models that detect similar products, and try to analyze probability of sales for each product. Analysis results can be used for planning future sales campaigns, marketing expenses optimization, creation of new loyalty programs, and better understanding customer behavior in general.