Demand Forecasting for Petrol Products in Gas Stations Using Clustering and Decision Tree

Abstract

Demand forecasting for petrol products in gas stations is crucial to the planning of initiative distribution of petrol products, especially to the stability of product supply in petroleum companies. In this paper, a novel scheme of demand forecasting based on clustering and a decision tree is proposed, which uses a decision tree and integrates the results of clustering validity indices. First, the proposed scheme uses a k-means algorithm to divide the sales data into multiple disjointed clusters, evaluates the clustering result of the daily sales curve of a product according to seven validity indices and determines the optimal number of clustering. Next, the relationship between the sales pattern and the relevant influence factors is described using a decision tree, which can categorize a future day’s sales pattern with these factors into the most suitable cluster to predict the quantity of the demand and the peak demand time windows for each gas station. Finally, three months’ worth of sales data is collected from a gas station in Dalian city, China, to illustrate the proposed forecasting scheme. Experimental results demonstrate that the scheme is an effective alternative for the demand forecasting for petrol products because it outperforms three other selected methods.