A Hybrid Subspace-Connectionist Data Mining Approach for Sales Forecasting in the Video Game Industry

This paper addresses the issue of sales forecasting using a new approach based on connectionist and subspace decomposition methods.A tool is designed to support company management in the process of determining expected sales figures. Neural networks trained with a back-propagation algorithm are used to predict the weekly sales of a video game. For this purpose, optimal topology is found and a time-sensitive neural network is implemented. We have considered the use of many influencing indicators and parameters as inputs. In order to assess the relevance of these parameters, we perform a pre-processing based on Principal Component Analysis. The performance of the proposed system is evaluated and compared with baseline reference sales. The results are presented and discussed with regards to prediction accuracy.