COMPARATIVE STUDY OF LSTM, ARIMA,AND PROPHET MODELS FOR STOCK MARKET TREND PREDICTION:A CASE STUDY ON GOLD PRICES

Abstract

Forecasting stock market trends is a complex task because financial data are highly volatile and often follow non linear patterns. Accurate forecasting models play an important role in helping investors, analysts, and financial institutions make informed decisions, reduce uncertainty, and manage investment risk more effectively. This study provides a comparative analysis of three widely used time series forecasting models, namely Long Short Term Memory (LSTM), Auto Regressive Integrated Moving Average (ARIMA), and Facebook Prophet, for predicting gold price trends. Historical gold price data from January 2021 to January 2024 were collected from Investing.com and used for model development and evaluation. After preprocessing, the dataset was divided into 80% training data and 20% testing data in order to preserve the chronological structure of the time series. Model performance was assessed using Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), R squared (R²), and Directional Accuracy. The results revealed that the LSTM model delivered the best overall performance, achieving the lowest MAE of 29.19, the lowest RMSE of 35.80, and the highest R² value of 0.62. These findings highlight the model’s strong ability to capture complex and non linear patterns in gold price movements. ARIMA demonstrated moderate overall performance but achieved the highest directional accuracy of 55.84%, showing its strength in short term trend direction prediction despite lower predictive precision. In contrast, Prophet produced the weakest results, with the highest error values and a negative R² score of 2.30, indicating limited effectiveness on highly volatile financial data. Overall, the findings suggest that LSTM is the most suitable model among the three for gold price forecasting, while ARIMA remains valuable for identifying directional movement. This study emphasizes the importance of choosing forecasting models based on both predictive accuracy and trend sensitivity, and it also highlights the potential of hybrid models and lightweight transformer-based approaches for improving real time financial forecasting in future research.