Research Article 2
- Volume 4, Issue 1 2026
By Chandar Kumar, Arjan Kumar, Ahmed Muddassir Khan, Muhammad Tayyab Yaqoob
10.20547/aibd.264103
Keywords: Machine learning; fuel price forecasting; ARIMA; Random Forest; Pakistan; crude oil; exchange rate; energy economics

The retail fuel price forecasting of emerging economies is important for the energy policy, fiscal planning and consumer welfare. Pakistan, where petrol and diesel prices are periodically revised (twice every fortnight) using an Import Parity Pricing (IPP) mechanism, is strongly influenced by changes in the international crude oil benchmarks (Brent and WTI) and Pakistani Rupee (PKR) to US dollar exchange rate. The paper presents a hybrid machine learning model that combines an autoregressive integrated moving average (ARIMA) model for trend extraction with a Random Forest (RF) regressor of nonlinear residual patterns. With real-time crude prices, historical PKR/USD weekly data (2026), and a realistically calibrated synthetic petrol price series based on data released by the Oil and Gas Regulatory Authority (OGRA), our model predicts the next price adjustment of Premium Euro 5 petrol. An empirical assessment of the price data on the April 30, 2026 (petrol: Rs 399.86/litre) demonstrates that the recent price spike was largely driven by a PKR depreciation to 380.17 on April 9, 2026. At the current price of Brent crude at 108.17 per barrel (as of May 2, 2026) and the PKR stabilizing at 368.34, the model predicts a reduction of PKR 21.97 per litre of petrol at the May 15, 2026 revision, with a 95% prediction interval, [372.78, 382.99]. The proposed framework is validated using the mean absolute percentage error (MAPE) of 0.62% during the period of January-April 2026, which is better than the standalone ARIMA model (MAPE 1.11%). Our results underscore the role of machine learning in converting market signals from a global perspective to local predictions of prices that can be implemented, a concept that applies to other import-dependent economies. Note: This is a historical price series that is synthetically generated to demonstrate the methodology, in practice this would be real OGRA data.

Submission Date: 7 Jan, 2026 Reviews Completed: 20 May, 2026
Acceptance Date: 8 Jun, 2026 Publication Date: 30 Jun, 2026