Sovereign Gold Bond Trend Line Prognostication using Multi-Headed Attention based Transformer

In order to forecast gold prices, a transformer-based deep learning method is presented in this work. The method uses a comprehensive data set that covers the years 2013–2023. Preparation of the data through normalization and temporal sequence structuring is part of the study’s early phases. The data set is then separated into training and test sets. The model is trained on the former, and its accuracy and Mean Absolute Percentage Error (MAPE) are used to evaluate it on the latter. The suggested transformer model is especially designed to represent the intricate temporal correlations found in time series data related to finance. It is made up of stacked multi-head self-attention layers, feedforward neural networks, and layer normalization, and it has several levels of self-attention processes. The Adam optimizer is used to train the models in 50 epochs with a batch size of 32. Experimental findings demonstrate the strong predictive ability of the model, with a test set accuracy of 0.9894 and a MAPE of 0.01055. These measurements show that the model outperforms conventional time-series forecasting methods in its ability to predict gold prices. Furthermore, the training process exhibits convergence, as evidenced by the training datas final Mean Squared Error (MSE) of about 0.0015, which shows successful learning. The transformer model’s capacity to identify intricate patterns and long-term interdependence in changes in the price of gold is further supported by a visual comparison of model predictions with actual prices and baseline models. These findings demonstrate its potential value in financial forecasting by offering insightful information about investment strategies and market trends.