Development of an Automated System for Predicting the Success of a Video Game Using Machine Learning

In the modern world game industry is constantly changing with new games being produced every day. Evaluating the sales volume of an application that has not yet been issued is essential for the developer. A program that can be used for this purpose was developed in the framework of this project. More than 16 thousand records on various videogames downloaded from the Internet served as source data. A benchmark model was built, based on the assumption of dependence between a game’s sales and critics’ and users’ reviews. Five machine learning models were chosen to be trained and tested on the data sample: Support Vector Regression, Decision Tree Regression, k Nearest Neighbours Regression, Gradient Boosting Regression Trees and Ridge Regression. As a result, the information was obtained on the optimal market areas for implementation of each of the machine learning models along with the information on the optimal model setting in order to achieve minimal error. Each of the models was compared to the benchmark model and to each other, providing a clear understanding of each model’s potential.