Application of Artificial Neural Network Models for Predicting Diesel and Petrol Prices in the Geographically Sparsed Regions in Tanzania

Abstract

Fuel consumption in Tanzania, mainly diesel and petrol, accounts for 82 percent of the energy consumption in the country, with significant price volatility affecting market stability, availability of fuel, and investment decisions. This study uses an artificial neural network (ANN) with a backpropagating algorithm to predict fuel prices in four regions of Tanzania. Key input parameters include the currency inflation rate (CIR), the petrol fuel inventory (PFI), the diesel fuel inventory (DFI), and the fuel transport costs (FTC). The study selected the 6-10-10-2 ANN structures for Sumbawanga-Rukwa, Mpanda-Katavi, and Mbeya-Mbeya as well as 6-10-9-2 for the Songea-Ruvuma region. The results show that transit distances between 200 and 400 km have a significant effect on the price of fuel, with petrol ranging from 0.1199 to 0.1349 Tanzania shillings per litre and diesel from 0.1203 to 0.1502 Tanzania shillings per litre. Road conditions also have an impact on fuel costs, with average fuel consumption of 0.9685 l/km on gravel roads versus 0.1325 l/km on paved roads. This finding suggests that poor road conditions contribute to higher fuel consumption and price volatility. Transport distances below 35 km have a minimal impact; however, load, speed, climate, and driving habits all contribute to variations. The results illustrate that the increase in distance influences higher price fluctuation for diesel than petrol. The study confirms that the application of ANN for predicting fuel price trends helps decision makers to make sustainable investments. The study recommends consolidation of transport and use of rail to reduce costs, although the limited rail network limits regional availability.