Abstract

An accurate forecast of stock market indices is a cornerstone of financial decision-making. This study employs a range of machine learning models to forecast the daily closing price of the Dow Jones Australia Index (DJ Australia) from 2015 to 2025. We comparatively evaluate the performance of eight regression models; Linear Regression, Support Vector Regression (SVR), XGBoost, Random Forest, k-Nearest Neighbors (KNN), Multi-layer Perceptron (MLP), LightGBM, and CatBoost by using a time-series split of the data. Feature engineering involved extracting temporal components (year, month, day, day of week) from the date. Model performance was assessed using Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE). Contrary to expectations that complex ensemble and deep learning models would dominate, the results indicate that Linear Regression outperformed all other models, achieving the lowest error metrics (RMSE: 29.853, MAPE: 7.05%). This surprising finding suggests that, for this specific dataset and features, the relationship between the temporal components and the index price is predominantly linear, or that more sophisticated models overfitted the training data. The results underscore the importance of model simplicity and baseline comparison in financial time series forecasting.