Gold Price Prediction Using Image Based ForCNN Model: A Deep Learning Approach

Abstract

The volatility of gold prices significantly influences global financial stability, necessitating the development of reliable models capable of producing precise forecasts to minimize investment risks and maximize profitability. Recently, both machine learning and deep learning approaches have gained significant traction for time series forecasting across scientific and industrial domains. In this paper, we propose the ForCNN model, which utilizes grayscale image-based input rather than traditional numerical data. This algorithm integrates the advantages of visual image representation of a time series and deep 2D convolution neural network to analyze and extract important features and generate accurate forecasts. We carried out extensive experiments on two real-world gold closing price datasets and showed ForCNN outperformed most of the state-of-the-art deep learning techniques such as MLP, CNN1D, LSTM, CNN-LSTM, BiLSTM, CNN-BiLSTM in terms of accuracy measures. Furthermore, portfolio performance evaluation using Cumulative Return, Average Daily Return, and Sharpe Ratio indicates that ForCNN achieves superior profitability and stronger risk-adjusted performance, thereby underscoring its effectiveness in practical financial forecasting applications.