Using Machine Learning Methods to Solve Problems of Forecasting the Amount and Probability of Purchase Based on E-Commerce Data

The study is aimed at investigating the possibility of using machine learning methods to build models for predicting the probability of purchase and the amount of purchase by online store customers. As a sample, we used data of users transactions of the site ponpare.jp in the period from 01.07.2011 to 23.06.2012. The description and comparative analysis of the most common methods for solving similar problems are given. The metrics used to measure the results in the case of forecasting the fact and amount of the purchase are being described. The results obtained make it clear that within the framework of the problem of predicting the probability of a purchase, gradient boosting, namely its implementation of LGBMClassifier, shows the most accurate estimate. For the problem of predicting the amount of a customer’s purchase, using gradient boosting also gave the best results.