CUSTOMER REVENUE PREDICTION FROM GEOGRAPHICAL DATA

Abstract

Online stores have created more opportunities for firms to offer different products and services to their customers. These online stores produce a tremendous amount of data that serve different purposes, including revenue prediction. Online stores usually keep a log of customers that visit their website that include session information, the products they showed interest, IP, location, and device information. These features are explored massively for studies such as customer churn and recommender systems but using location information for prediction is not explored as much.

The first part of this thesis systematically reviews the articles on revenue prediction with respect to their publication date, application area, evaluation criteria, and technique for prediction that provides a good understanding of already conducted research, the evolution of the topic over the years, and possible research opportunities.

The second part focuses on the prediction of Google store revenue data. Using linear regression as a baseline, it evaluates the predictive power of different machine learning techniques, including gradient boosting, support vector regression, and neural networks. The data is collected from Google Analytic demo account that contains 903,653 observation and 55 features. The goal of this study is to predict the total transaction per user from December 1st, 2018 to January 31st, 2019 and in order conduct performance analysis between different prediction techniques.