Clustering-based Sales Forecasting in a Forklift Distributor

Abstract

Sales forecasting refers to the prediction of future demand based on past data. A vast literature on sales forecasting has accumulated due to its vital role in balancing demand and supply. Among these, data mining has emerged as a powerful tool to facilitate sales forecasting. In this study, we use data mining methods for accurate and reliable sales forecasts in a forklift distributor company. Monthly sales data for 100 different types of forklifts between 1998 and 2016 are used. The proposed forecasting methodology includes three steps. First, products with similar sales patterns are determined using hierarchical clustering. Dynamic time warping is applied to calculate the similarities among product sales data. Second, features are extracted and selected for each cluster. In addition to the features adopted from the literature, four new features are proposed to characterize intermittency. Multivariate adaptive regression splines model is used for feature selection. Third, support vector regression is used to predict future sales of each product cluster. Finally, the performance of the proposed approach is evaluated according to forecasting error and complexity. The numerical analysis shows that the proposed approach gives reasonable accuracy with less complexity.