Overview Gold price prediction is crucial for financial markets, and this project explores the effectiveness of LSTM networks for forecasting gold prices. It compares models trained to predict short-term and long-term trends, helping to identify the most reliable model for different forecasting windows.
Key Features Short-Term vs. Long-Term Forecasting: The project contrasts models trained for short-term and long-term predictions. LSTM Model Architecture: Utilizes LSTM networks for time series forecasting, known for their ability to capture sequential dependencies. Data Processing: Preprocessing of time series data for feature extraction, normalization, and splitting into training/testing datasets. Evaluation: Performance of the models is evaluated using common metrics like Mean Squared Error (MSE) and Root Mean Squared Error (RMSE).
Results:
Short-Term Model: Forecast Horizon: Predicting gold prices for the next 7 days.
Performance Metrics:
Root Mean Squared Error (RMSE): 19.46
Mean Absolute Percentage Error (MAPE): 0.37%
Visualization: Predicted gold prices closely followed the trend of actual prices, demonstrating that the LSTM model effectively captured short-term fluctuations in gold prices.
picture credit: TradingView. (n.d.). TradingView — Track all markets. https://www.tradingview.com/
Long-Term Model: Forecast Horizon: Predicting gold prices for a longer period (30-60 days). Performance Metrics: RMSE: Higher compared to the short-term model, indicating increased prediction error over the longer horizon. MAPE: Slightly higher than the short-term model, reflecting the challenge of forecasting over extended periods.