Please use this identifier to cite or link to this item:
Title: Precious metal price prediction using deep neural networks
Authors: Peng, Zhiling
Keywords: Engineering::Electrical and electronic engineering
Issue Date: 2021
Publisher: Nanyang Technological University
Source: Peng, Z. (2021). Precious metal price prediction using deep neural networks. Master's thesis, Nanyang Technological University, Singapore.
Abstract: Gold has always played an essential role in financial activities as it is treated as a global currency, an attractive investment and is the raw material in many industries. Because of the multiple attributes of gold, the price of it will be affected by a multitude of factors, including the supply and demand, global economic conditions, national policies, and many other factors such as the U.S. dollar index, the U.S. dollar exchange rate, crude oil prices, etc. Therefore, how to predict the trend of the gold price effectively has been long concerned by researchers from many areas, and such prediction can be really helpful to control the risk of personal investment as well as adjust the national economic strategies in time. Nowadays, the mainstream gold price forecasting methods are machine learning methods, among which many regression models and the deep learning algorithms are widely accepted. In this project, the predicting model proposed by Vidya and Hari [1] is reproduced, using the long short-term memory (LSTM) network model. The results we obtain are roughly the same as those observed by Vidya and Hari. Then, four traditional regression methods in machine learning are used to train and evaluate the same dataset of the gold price, including the linear regression, support vector regression (SVR), decision tree regression and the random forest algorithm. The results show that the multiple linear regression method even outperforms the LSTM model proposed by Vidya and Hari [1]. However, traditional regression methods have limited performances when processing long sequences and are prone to have underfitting or overfitting problems. In order to solve the problems of the traditional regression methods, a variety of neural network models are constructed for gold price prediction, among which the convolutional neural network (CNN), the LSTM network and their combinations (the CNN-LSTM model and the LSTM-CNN model) are used. The results indicate that the LSTM-CNN model is effective for the long-sequence gold price prediction and the performance is better than the LSTM model. In addition, in order to consider the potential correlations among the input data in two directions, the bidirectional LSTM model (Bi-LSTM) is proposed and is also combined with the CNN network (the CNN-Bi-LSTM model and the Bi-LSTM-CNN model), thus improving the predicting accuracy and achieving better results. Finally, other factors such as the price of the crude oil and the U.S. dollar index are taken into consideration. Although the predicting accuracy is not improved with multi-source data used, it also provides new ideas for the gold price forecasting. Among all these approaches used in this project, the Bi-LSTM-CNN model has the best performance with the RMSE of 6.604, having an improvement of 10.6% compared to the testing RMSE of 7.385 obtained by Vidya and Hari [1]. Keywords: gold price prediction, deep learning methods, regression models, the LSTM network, the Bi-LSTM model, multiple factors
Fulltext Permission: restricted
Fulltext Availability: With Fulltext
Appears in Collections:EEE Theses

Files in This Item:
File Description SizeFormat 
Precious Metal Price Prediction Using Deep Neural Networks.pdf
  Restricted Access
1.69 MBAdobe PDFView/Open

Page view(s)

Updated on May 15, 2022


Updated on May 15, 2022

Google ScholarTM


Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.