GASOLINE CONSUMPTION PREDICTION VIA DATA MINING TECHNIQUE

Abstract

Due to the increasing dependence of human life on energy, it plays a crucial role in the functioning of the various economic sectors of the countries, potentially and actually. Fuel products, especially gasoline, given their importance in the transportation sector, play major roles in the economic growth and development of countries. Hence, the authorities in each country have to control the fuel supply and demand parameters accurately with a more accurate prediction of fuel consumption and proper planning in the direction of consumption. The purpose of this study is to find appropriate methods and approaches for forecasting gasoline consumption in Tehran using data mining methods. For this purpose, daily consumption data of gasoline stations were collected in 5 different regions of Tehran during the period of 2008-2013. Then, these numbers were predicted on a daily, weekly, monthly, and seasonal basis for analyzing the consumption at different time intervals. The standardization method was also used to match the scales. After data pre-processing, gasoline consumption was predicted using the multi-layer perceptron (MLP) neural network method. The gasoline consumption forecast was evaluated based on the mean squared error (MSE), mean, and mean absolute error (MAE) criteria. The results indicate that the artificial neural network (ANN) can accurately predict gasoline consumption in five different regions of Tehran.