Precious Metal Prices Forecasting Using Optimally Configured Hybrid Deep Learning Approach
Jumana Waleed, Taha Mohammed Hasan, Ala'a Jalal Abdullah, Ahmed Hussein Alkhayyat · Journal of Machine and Computing · 2025
Precious metals price forecasting represents an intricate task owing to their elevated volatility and delicacy to global economic variations. Conventional time series forecasting approaches frequently attempt to account for the non-linear and complex relationships that exist in commodity price movements, resulting in sub-optimal accuracy in price forecasting. Recently, the emergence of deep learning has provided outstanding modeling of such intricate patterns. This paper investigates the implementation of deep learning approaches, particularly One Dimensional Convolutional Neural Networks (1D-CNN), Long Short-Term Memory (LSTM), and the combination of 1D-CNN and LSTM, for precious metals prices forecasting. By drawing on the competitive unique capabilities of 1D-CNN in extracting essential features, LSTM in sequential data processing, and Hyperband optimization methodology in automatically optimizing hyper-parameters, the proposed hybrid approach endeavors to improve forecasting accuracy compared to individual approaches. Extensive experiments are conducted to assess the performance of implemented approaches using three datasets traded at the Multi Commodity Exchange (MCX), and the attained accuracy exhibits the hybrid approach’s superiority over standalone architectures. Using the gold dataset as an example of a precious metal, the proposed hybrid approach results for the Absolute Error (MAE), Root Mean Squared Error (RMSE), and Rsquared were 0.0182, 0.1500, and 0.9616, respectively. The outcomes indicate that the proposed hybrid forecasting approach of 1D-CNN and LSTM can considerably enhance the capabilities of prediction in the precious metal price forecasting field, providing an encouraging architecture for analyzing the financial market.