@Article{iasc.2026.078344,
AUTHOR = {Thanh Tuan Nguyen, Cuong Nguyen Dinh Hoa},
TITLE = {Local Feature Extraction and Time-Series Forecasting of Crude Oil Prices Using 1D-CNN},
JOURNAL = {Intelligent Automation \& Soft Computing},
VOLUME = {41},
YEAR = {2026},
NUMBER = {1},
PAGES = {1--24},
URL = {http://www.techscience.com/iasc/v41n1/67355},
ISSN = {2326-005X},
ABSTRACT = {Accurate crude oil price forecasting is critical for global economic stability but remains an exceptionally challenging task due to the data’s complex, non-linear, and non-stationary nature. Deep learning models like LSTMs are widely favored. However, the dominant research trend currently focuses on increasingly complex hybrid and ensemble architectures. These models often suffer from high computational overhead, intricate tuning processes, and potential overfitting, raising critical questions about their necessity. In this paper, we challenged the assumption that complexity is required for high performance by proposing and evaluating a streamlined 1D-CNN model. We conducted a comprehensive evaluation of this standalone architecture against a standard LSTM network, a hybrid 1D-CNN-LSTM model, and a Naive Persistence baseline. The experimental evaluation was performed across three distinct forecasting scenarios: one single-step and two multi-step prediction tasks. Our quantitative results showed that the proposed 1D-CNN model consistently and decisively outperformed both baselines across all three scenarios. The 1D-CNN achieved the lowest MAE, MSE, and RMSE, and the highest <mml:math id="mml-ieqn-1"><mml:msup><mml:mi>R</mml:mi><mml:mrow><mml:mn>2</mml:mn></mml:mrow></mml:msup></mml:math> score, with qualitative analysis confirming its superior predictive alignment. This work demonstrates that a simpler, standalone 1D-CNN architecture provides a more effective and efficient solution for crude oil price forecasting, challenging the prevailing trend toward escalating model complexity.},
DOI = {10.32604/iasc.2026.078344}
}