Optimal Gasoline Price Predictions Using ANFIS Model

This project implements an Adaptive Neuro-Fuzzy Inference System (ANFIS) model to predict optimal gasoline prices based on historical data and relevant features.

Overview

Gasoline prices fluctuate due to various economic and geopolitical factors. Accurate price prediction models help stakeholders make better decisions. This project applies an ANFIS model combining neural networks and fuzzy logic to forecast gasoline prices with improved accuracy.

The approach is inspired by and adapted from the methodology in the paper: "Optimal Gasoline Price Predictions: Leveraging the ANFIS Regression Model" DOI: 10.1155/2024/8462056

The work is implemented as a Kaggle notebook available here.

Dataset

The dataset used in this study is sourced from the U.S. Energy Information Administration (EIA) and contains weekly U.S. retail gasoline prices (in dollars per gallon) across all grades and formulations.

🗓️ Time Span

• Start Date: April 5, 1993
• End Date: July 31, 2023
• Total Samples: 1,583 weekly observations

📈 Data Characteristics

• Frequency: Weekly (one observation per week)
• Unit: U.S. Dollars per Gallon
• Type: Univariate time series (only price data considered)
• Coverage: Nationwide U.S. average prices encompassing all gasoline grades and formulations

This dataset captures both long-term historical trends and short-term fluctuations in gasoline prices, making it well-suited for time series modeling and predictive analytics.

📎 Note

The dataset is publicly available from the U.S. EIA website and can be accessed here: https://www.eia.gov/dnav/pet/hist/LeafHandler.ashx?n=PET&s=EMM_EPM0_PTE_NUS_DPG&f=W

Features

• Feature engineering and selection based on domain knowledge
• Data preprocessing, including normalisation and cleaning
• Implementation of ANFIS for price prediction

Model

• Adaptive Neuro-Fuzzy Inference System (ANFIS)
• Hybrid learning approach combining gradient descent and least squares
• Evaluation using metrics such as RMSE, MAE, etc.

Usage

1. Clone this repository:

   git clone https://github.com/gamzeakkurt/anfis-gasoline-forecasting.git
   cd anfis-gasoline-forecasting

2. Install required packages:

   pip install -r requirements.txt

3. Run the notebook or Python scripts to train and evaluate the ANFIS model.

Results

• The model achieves competitive performance on the test set
• Visualisations of predictions vs actual prices included

References

• Jang, J.-S.R. (1993). ANFIS: Adaptive-Network-based Fuzzy Inference System.
• Çakır, M., & Sağlam, M. (2024). Adaptive neuro-fuzzy inference system for optimal gasoline price prediction. Journal of Applied Mathematics and Computation, 2024, Article ID 8462056. https://doi.org/10.1155/2024/8462056