A Novel Hybrid Approach to Forecasting Crude Oil Prices Using Local Mean Decomposition, ARIMA, and XGBoost

Crude oil in its usage and process brings about several issues that affect the environment including soiling of the air and water and the release of greenhouse gases leading to changes in the climate. Further, historically use of oil has created political and economic stakes that primarily end up in war, nationalism to control resources, and political instability. As a new approach to this problem in this study, we present a novel hybrid of Local Mean Decomposition (LMD) for breaking down the data and contribution coefficients for rebuilding the component with less computational complexity. Using the XGBOOST approach, deep dynamic learning features, augmented with autoregressive moving average (ARIMA), it is possible to predict each reconstructed segment. Turbulences that occur in the oil markets are random, the market shares are sensitive to pressures from other markets, the movement in the market is disorderly, and there is high volatility in the market. Thus, the price of oil is very unfixed, unpredictable, sensitive, and uncertain. Several techniques have been adopted in cost estimation in the aspect of time variation; some of the traditional methods include EMD, EEMD couldn't solve the computational complexity and save time and cost. Because of the unpredictability and fluctuation observed in oil prices non deep neural network models and data mining cannot be used in the prediction of costs and generalized forecasting. The analysis of the onset of deep non-linear modeling as a part of the machine learning process demonstrates how the discussed hybrid model successfully acquires relevant data and adapts a non-stationary function. The data is modeled using an LMD-SD-ARIMA-XGBOOST hybrid recurrent network when the nonlinearity is considered, and the residual values are forwarded to the XGBOOST model. Besides the volatility problem, the solution provided by the work also addresses the overfitting problem of neural networks. The hybrid LMD-SD-ARIMA-XGBOOST model overcomes overfitting by combining statistical and machine learning components, where the decomposition process isolates key features of the data, reducing noise and complexity. Unlike standalone neural networks, which require large amounts of data to generalize effectively, our hybrid approach leverages ARIMA and XGBOOST models that are inherently less prone to overfitting due to their simpler structures and feature-focused learning. To validate the proposed hybrid approach, data from the West Texas Intermediate (WTI) which is publicly available is employed.