Forecasting Gold and Platinum prices with an enhanced GRU model using multi-headed attention and skip connection

Metal price prediction is a critical task for policymakers, financial analysts and investors because of its significant impact on global markets and economies. This study investigates the effects of precious metal prices including gold and platinum with a multivariate model. In this work, we propose a novel multi-headed attention based GRU with skip connection (MA-GRUS) for metal price prediction. It consists of various steps including data preprocessing, model building, training, evaluation and visualization. The Gated Recurrent Unit (GRU) model is built with a multi-layer architecture to find temporal dependencies while multi-headed attention mechanisms along with skip connection are used to improve the model’s ability to find relevant and significant information. We evaluate the efficiency of the proposed model with the help of Mean Absolute Error (MAE), Mean Squared Error (MSE), Root Mean Squared Error (RMSE), R2 and training time providing a comprehensive assessment of its predictive accuracy. The obtained results show that the proposed model outperformed over Long Short-Term Memory (LSTM), Recurrent Neural Network (RNN), GRU, Bi-directional GRU, and attention based GRU in terms of prediction accuracy and error rate. The performance of GRU is enhanced with the help of multi-headed attention mechanism and skip connection. The model is simpler to implement and performs rapid training because of its simple architecture and layers details. It contributes to the field of time series forecasting by leveraging powerful efficiency of GRU to capture complex patterns in financial data.