Two Layered Deep Long Short-Term Memory Network based Gold Rate Divination

Abstract

From the moment of its discovery, gold has been the most significant and valuable commodity. The exact forecasting of gold price in the market remains a major challenge for the researchers to focus upon. With this motivation, this paper propose a novel Dual Coping Deep Long Short-Term Memory Network to forecast the gold price with minimum mean squared error. The model utilize the Daily gold price historical dataset retrieved from the KAGGLE machine-learning repository. The Daily gold price historical dataset contains 7 features with 5837 gold price details and is processed with missing values and encoding. The exploratory data analysis have done to visualize how the independent features are related to the gold price. The proposed Dual Coping Deep Long Short-Term Memory (DC-DLSTM) Network had designed with two LSTM overlay that recognizes short- and long-term relationships and convolutional layers retrieves the meaningful information from time-series gold price data and comprehending its internal representation. The Daily gold price historical data have applied to proposed model, same data was applied to other regressor models, and the performance metrics was analyzed. Implementation was scripted using python through NVidia Tesla V100 GPU workstation with 30 training epochs and batch size of 64. The performance metrics utilized for analysis are Mean squared error. Experimental results shows that the proposed DC-DLSTM shows the least mean squared error of 0.0013 when compared with other regressor models.