Authors:
Deebom Zorle Dum
Addresses:
Department of Mathematics, Rivers State University, Port Harcourt, Rivers State, Nigeria.
Abstract:
AI-Driven Forecasting of Nigeria's Daily Crude Oil Prices Utilizing the Integrated ARLAS Framework: Hybrid Statistical–Machine–Deep Learning Model examines how linear models fail to predict Nigeria's highly unstable and nonlinear crude oil price fluctuations. After cleaning and time series conversion, the dataset contains 2,968 daily crude oil prices from October 23, 2009, to September 30, 2024. Exploratory Data Analysis (EDA) and the Augmented Dickey-Fuller (ADF) test confirmed non-stationarity, which was addressed. The ARLAS Framework hybridises ARIMA with LSTM, ANN, and SVR to model linear and nonlinear relationships using the AIC-selected ARIMA (1,1,1) model. ARIMA captures linear trends effectively, but its out-of-sample forecasting accuracy is poor, with an RMSE of 14.46 and an MAE of 12.40. By identifying nonlinear trends and volatility patterns, the ANN (RMSE = 3.17, MAE = 1.81) and SVR (RMSE = 2.95, MAE = 1.39) are more predictive. The hybrid ARLAS model enhances forecast reliability and reduces residual errors, as shown by graphical and diagnostic assessments. Statistics, machine learning, and deep learning technologies make the combined ARLAS Framework better than ARIMA, according to the study. Hybrid AI-driven solutions for forecasting oil prices, formulating policies, and maintaining energy market stability in Nigeria are recommended for their ability to manage complex, non-stationary, and high-frequency financial datasets.
Keywords: AI-Driven Forecasting; Exploratory Data Analysis; Autoregressive Integrated Moving Average; Long Short-Term Memory; Artificial Neural Network; Support Vector Regression.
Received: 11/11/2024, Revised: 14/01/2025, Accepted: 22/03/2025, Published: 09/09/2025
DOI: 10.64091/ATITP.2025.000158
AVE Trends in Intelligent Technoprise Letters, 2025 Vol. 2 No. 3 , Pages: 143-156