Newspaper demand prediction and replacement model based on fuzzy clustering and rules

Abstract

A problem that most newspaper companies encounter daily is how to predict the right number of newspapers to print and distribute among distinct selling points. The aim is to predict newspaper demand as accurately as possible to meet customer need and decrease loss, the number of newspaper offered but not sold. The right amount depends of the newspaper demand at different selling points and is a function of the geographical location and customer profile. Currently, demand prediction is based on values experienced in the past and on management knowledge. This paper suggests the use of predictive data mining techniques as a systematic approach to explore newspaper company database and improve predictions. Predictions require accurate forecast of the daily newspaper amount needed at each selling point. The focus of the paper is on a prediction method that uses fuzzy clustering for data base exploration and fuzzy rules together with performance scores of selling points for prediction. Experimental results using actual data show that the method is effective when compared with the current methodology, neural network-based predictors, and autoregressive forecasters. In particular, the predictive data mining technique improves on average 10% in comparison with the use of the existing approaches.