Deep learning approach for forecasting silver prices: BILSTM

Deep learning is one of the most studied and popular topics today. With the advancement of technology and the facilitation of access to labeled data, it has become more focused on it. Deep learning is a sub-branch of the multi-layered Artificial Neural Networks structure of the artificial intelligence that works by learning through labeled data. Bidirectional Long Short-Term Memory (BILSTM) is one of the methods included in deep learning and has become popular recently. It can also be said that BILSTM is a modified version of the Long Short-Term Memory (LSTM). It has two separate hidden layers that is backward and forward to capture past and future information. These two layers are then combined to create the final output. The final output in question includes forecast or estimation results. In this study, a forecasting algorithm was established with the BILSTM model using daily data of silver (gr/TL) between 01.11.2017-30.09.2024. Five independent variables (dollar price ($/TL), consumer price index (%), housing price index, gold price (gr/TL), bitcoin price (TL)) that are thought to affect the price of silver are included in the model. In addition to the BILSTM model, LSTM and CNN-LSTM models were also established in the study and the success of each model was compared with each other through the RMSE success metric. The RMSE value of the BILSTM model was 0.8391, the RMSE value of the LSTM model was 1.7781 and the RMSE value of the CNN-LSTM model was 1.4312. Considering the RMSE value of each model, it was observed that the BILSTM model made a more successful prediction. CNN-LSTM model predicted poorly compared to BILSTM model and successfully compared to LSTM model. The LSTM model predicted poorly compared to the other two models.