Gold price prediction using long short-term memory neural network

Abstract

Gold has always been valued throughout human history, playing a significant impact on the economy. There is a need to monitor gold prices as they are seen as good indicators for the economic market. The ability to predict future gold prices for investors and governments is important, as it helps them to make better investments and decisions for both the economy and the country. In recent years, the advancement of technology alongside machine learning has led to artificial intelligence(AI) that has superseded human capabilities. A subset of machine learning, called deep learning, allowed for the creation of newer neural networks and AI models such as Long Short-Term Memory (LSTM) that has seen success in time-series data prediction. This project uses inputs such as the historical price of gold, silver and crude oil in an LSTM model to effectively predict the price of gold. The model is also further subjected to variations in parameters to find the best possible setting which returns the most accurate predictions. This provided the parameters of 32 neurons in 2 hidden layers, a dropout rate of 0.2 in the first layer and 0.3 in the second layer, a window size of 80, a batch size of 16, 1000 epochs, and RMSProp optimizer for most ideal prediction.