Gold Price Prediction Based On Long Short-Term Memory (LSTM) For Investment Decision-Making

Abstract

The movement of gold prices serves as a critical indicator in investment decision-making, especially in dynamic and uncertain market conditions. This study aims to develop a gold price prediction model based on Long Short-Term Memory (LSTM), a type of Recurrent Neural Network (RNN) capable of capturing temporal patterns in historical price data. The goal of implementing this model is to generate more accurate predictions compared to conventional methods, thereby supporting more informed investment decisions. The research utilizes daily gold price data over a specific period, combined with other economic indicators such as oil prices, exchange rates, and interest rates. The data undergoes preprocessing steps, including normalization and division into training, validation, and testing sets. The proposed model architecture consists of a single LSTM layer with 64 neurons and an output layer, trained over 50 epochs with a batch size of 32. The results show that the LSTM model achieves high prediction accuracy. This is demonstrated by evaluation metrics including a Mean Absolute Error (MAE) of 26.12, Mean Squared Error (MSE) of 1269.15, Root Mean Squared Error (RMSE) of 35.63, and an R² score of 0.9858, indicating that the model can explain 98.58% of the variance in actual gold price data. Furthermore, the visualization of the results indicates that the model is capable of closely following gold price trends and consistently predicting upward price movements over the next 30 days. The model has also been successfully implemented into a web-based platform using TensorFlow.js, enabling users to access real-time predictions in an efficient and responsive manner.