{"paper_id":"267ea82c-3243-4d6c-a43f-e6a4785de01a","body_text":"Forecasting the Dow Jones Australia Index: A Comparative Evaluation of Machine Learning Regression Models | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Forecasting the Dow Jones Australia Index: A Comparative Evaluation of Machine Learning Regression Models Sevda Kuşkaya, Faik Bilgili This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7473138/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version 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. Artificial Intelligence and Machine Learning Econometrics Macroeconomics Applied Statistics Finance Computational Mathematics Stock Market Prediction Machine Learning Time Series Forecasting Linear Regression Model Comparison Dow Jones Australia Index Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 1. Introduction The inherent volatility and complex, non-linear nature of stock markets make their prediction a challenging yet highly sought-after goal for investors, analysts, and economists. The ability to forecast market movements, even with marginal accuracy, can provide a significant advantage. The Dow Jones Australia Index, a key indicator of the Australian equity market's health, is influenced by a multitude of factors, including global economic trends, commodity prices, and domestic policy changes. With the advent of powerful computational resources and sophisticated algorithms, machine learning (ML) has emerged as a promising tool for uncovering hidden patterns in financial data that traditional statistical methods might miss. Models like Random Forests, Gradient Boosting Machines (e.g., XGBoost, LightGBM, and CatBoost), and neural networks are increasingly being applied to forecast financial time series. This study aims to contribute to this field by conducting a comprehensive empirical evaluation of multiple machine learning regression models for predicting the daily closing price of the DJ Australia index. The primary objective is to identify the model that provides the most accurate forecasts based on a set of engineered temporal features. The surprising outcome, which saw a simple Linear Regression model outperform more complex alternatives, provides a critical insight into the nature of this particular forecasting problem and highlights a crucial lesson in model selection. 2. Literature Review The application of machine learning to stock price prediction is a well-established domain within computational finance. Early work often focused on traditional time series models like ARIMA (Auto Regressive Integrated Moving Average). However, their limitations in capturing non-linear relationships and complex market dynamics led researchers to explore ML techniques Numerous studies have investigated the efficacy of various models. Support Vector Machines (SVM), particularly SVR for regression tasks, have been used for forecasting due to their effectiveness in high-dimensional spaces (Gandhmal & Kumar, 2019 ). Recently, ensemble methods have gained prominence. Random Forests, introduced by Breiman (2001), are praised for their robustness against overfitting. Gradient boosting frameworks like XGBoost (Chen & Guestrin, 2016 ) and LightGBM (Ke et al., 2017 ) have consistently delivered state-of-the-art results in various Kaggle competitions, including those involving financial data, due to their speed and performance. Deep learning approaches, such as Multi-layer Perceptrons (MLP) and Recurrent Neural Networks (RNNs), have also been applied to model temporal dependencies in sequential data like stock prices (Hoseinzade & Haratizadeh, 2019 ). Despite the sophistication of these models, the literature also documents instances where simpler models can perform on par with or even outperform complex ones, especially when the feature set is limited or the underlying relationship is linear. Dixon et al. ( 2020 ) provide a broad overview of ML in finance and discuss the practical considerations and pitfalls, such as the fact that simpler models can be more robust and effective. This body of work validates a comparative approach but also cautions against assuming that superior complexity always yields superior results. Vashistha ( 2025 ) underlines that ML can predict whether a particular stock trade will be profitable or not and shows how to predict a signal of stock prices. Kuşkaya and Bilgili ( 2025 ) found that among 19 ML models, the Linear Regression and CatBoost Regressor models achieved the best results in terms of accuracy and computational efficiency. Additionally, they noted that the Linear Regression model using the TimeSeriesSplit strategy demonstrated high ex-post and ex-ante forecast accuracy for the US electricity prices index. Our study adds to this literature by providing a direct, head-to-head comparison of these diverse model families on the DJ Australia index, using a minimalist feature set based solely on time, and by explicitly documenting a case where parsimony prevails. 3. Data The data for this study were sourced from a CSV/Excel file of DJ Australia (Investing.com) containing daily historical data for the Dow Jones Australia Index (USD). The original dataset comprised 2,590 entries (rows) spanning from July 28, 2015, to June 30, 2025. Each entry included the following financial variables: Date: The trading day (recorded as an object and later converted to datetime). Price: The daily closing price of the index (the target variable for prediction). Open: The daily opening price. High: The highest price reached during the day. Low: The lowest price reached during the day. Change %: The percentage change from the previous day's closing price. A preliminary exploratory data analysis (EDA) was conducted. The data was checked for missing values, and one null entry in the 'Price' column was identified and subsequently removed to ensure data integrity. Distribution plots (histograms) and boxplots were generated for all numerical variables to understand their spread and identify potential outliers. The initial time series plot of the closing price (Fig. 1) confirmed a clear upward trend over the entire period, which is a common characteristic of major equity indices over the long term. Figure 1 presents the daily closing price (Price) of the Dow Jones Australia Index from July 2015 to June 2025. The x-axis represents time (Date), and the y-axis represents the price in USD. The plot reveals a strong, persistent upward trend over the entire decade, which is a common characteristic of major equity indices over the long term, reflecting economic growth and market expansion. Except for one business day observation on March 23, 2020 (pandemic crash), the graph shows a relatively steady growth period for the Australian market during this time. This visual establishes the non-stationary nature of the data, a key challenge in time series forecasting. This panel of histograms shows the frequency distribution of the key financial variables in the dataset: Price, Open, High, Low, and Change %. The most critical observation is the bimodal distribution (two distinct peaks) for all four variables. This strongly suggests that the market exhibited two different \"regimes\" or phases during the 2015–2025 period: The first peak of the price regime (lower price regime) likely corresponds to the earlier years in the dataset, such as 2015–2019. A higher-price regime (second peak) likely refers to the later years (e.g., post-2020 recovery and subsequent growth). This multimodality indicates a significant structural shift in the market, potentially driven by major economic events (e.g., the post-pandemic recovery). This insight is crucial because it challenges the assumption of a single, stable distribution in many forecasting models. Figure 3 of Box Plots of Financial Features visualizes the spread and central tendency of the same financial features and identifies potential outliers. The box represents the interquartile range (IQR-the middle 50% of the data), the line inside the box is the median, and the ‘whiskers’ extend to show the range of typical data points. Points beyond the whiskers are considered outliers. Given the strong correlation between these values on any given day, the Price, Open, High, and Low plots are nearly identical. They show a positive skew (the median is closer to the bottom of the box, and the upper whisker is longer), consistent with the upward trend seen in Fig. 1. There are a few extreme outliers. The change % plot is centered around zero, with a symmetrical distribution. The presence of numerous outliers on both the positive and negative sides is normal for daily percentage change data, reflecting days of unusually large gains or losses. 4. Methodology The overall methodology followed a standard supervised machine learning workflow for regression, tailored for time-series data to avoid look-ahead bias. 4.1. Data Preprocessing and Feature Engineering: The 'Date' column was converted into a datetime object and used as the basis for sorting the entire dataset chronologically in ascending order. This is a critical step for time-series analysis. The index was reset to ensure proper alignment for subsequent operations and plotting. Instead of using the raw financial features (Open, High, Low, Change %), which were discarded to avoid leakage and simplify the initial model comparison, the Date was decomposed into numerical features. This feature engineering step extracted the following explanatory variables: Year, Month, Day, and Day-of-Week (Monday = 0, Sunday = 6). This transformed the problem into predicting the index price based solely on temporal components. 4.2. Train-Test Split: To preserve the temporal order of the data and simulate a realistic forecasting scenario, a sequential split was used instead of a random shuffle. The first 80% of the data (2015–2023) was used for training, and the last 20% (2023–2025) was used for testing. This ensures that models are trained on past data and evaluated on future, unseen data. 4.3. Model Selection and Training: A diverse set of eight regression algorithms was selected to represent different modeling paradigms: Linear Models : Linear Regression Support Vector Machines : Support Vector Regression (SVR) Ensemble Methods : Random Forest (RF), XGBoost (XGB), LightGBM, CatBoost Instance-Based Learning : k-Nearest Neighbors (KNN) Neural Networks : Multi-layer Perceptron (MLP) All models were trained on the same feature set (Year, Month, Day, Day-of-Week) and target variable (Price) using their default hyperparameters in Scikit-Learn to provide a baseline comparison of their out-of-the-box performance. 4.4. Performance Evaluation: Model performance was evaluated on the test set using a comprehensive set of metrics to assess different aspects of prediction error: Root Mean Squared Error (RMSE) : It prioritizes larger errors, useful for understanding typical error magnitude in the units of the target variable. Mean Absolute Error (MAE) : It provides a linear score for the average error magnitude. Mean Absolute Percentage Error (MAPE) : It expresses error as a percentage, making it easily interpretable. The models were ranked primarily by RMSE for final comparison. The results were compiled into a comparative table and visualized using bar charts for error metrics and a time series plot comparing the actual values to the predictions of the best- and worst-performing models. Table 1 ML Model Comparison Results Model RMSE MAE MAPE 1 Linear 29.853786 23.2086 7.054014 2 KNN 58.521712 50.953445 15.257116 3 CatBoost 58.554258 50.946156 15.254353 4 XGB 59.847303 52.40042 15.665139 5 RF 59.886671 52.149562 15.605581 6 LightGBM 60.150836 53.04236 15.843552 7 SVR 83.57932 78.942095 23.261142 8 MLP 85.142504 80.620807 23.739671 In Figs. 4 and 5, the bar charts provide a visual comparison of the forecasting errors for the eight tested models, ranked from best (Linear) to worst (MLP). Figure 4 compares three error metrics (forecast/prediction accuracy metrics) simultaneously: RMSE, MAE, and MAPE. The dramatic difference in the bar for the Linear Regression model across all three metrics immediately conveys its superior performance. Figure 5 reveals the MAPE (Mean Absolute Percentage Error). This is often the most intuitive metric, as it expresses the error as a percentage. It clearly shows that Linear Regression's average error is around 7%, while all other models have an error of approximately 15% or higher. This chart effectively communicates that the simple linear model is, on average, more than twice as accurate as the complex models for this specific task. In Fig. 6, the Actual vs. Predicted Prices time series plot is perhaps the most intuitive illustration of model performance. It shows the actual values of the index in the test set (the holdout period from 2023 to 2025) compared to the predictions made by the best model (Linear Regression) and the worst model (Multi-layer Perceptron - MLP). The black line (Actual) shows the true market movement. The dashed line (Linear Prediction) closely follows the black line's general direction and turning points. While it doesn't capture every minor fluctuation, it accurately tracks the overall trend and rhythm of the market. The dotted line (MLP Prediction) is highly erratic and fails to follow the actual data meaningfully. It demonstrates severe overfitting; the complex MLP model learned noise from the training data instead of the underlying pattern, making its predictions on new, unseen data (the test set) essentially useless. This visual starkly contrasts a model that generalized well (Linear) with one that failed (MLP), powerfully reinforcing the paper's central finding 5. Results The dataset consisted of 2,590 daily records from July 2015 to June 2025. After sorting by date and splitting sequentially (80% train, 20% test), the training set contained 2,071 samples and the test set 519 samples. Eight regression models were trained and evaluated on the test set. The performance metrics for all models are summarized in Table 1. The models are ranked from best to worst based on RMSE. Best Performing Model Linear Regression was the clear top performer. It achieved an RMSE of 29.853, an MAE of 23.208, and a MAPE of 7.05%. Intermediate Performers KNN, CatBoost, XGBoost, and Random Forest formed a middle group. Their error metrics were roughly twice as high as those of Linear Regression (RMSE ~ 58.52–59.89, MAPE ~ 15.26–15.60%). Worst Performing Models SVR and MLP were the least accurate models, with the highest errors (RMSE > 83, MAPE > 23%), indicating they performed worse than a simple horizontal line predicting the mean. Figures 4 and 5 (metrics comparison bar charts) visually confirm the significant advantage of the Linear model. Figure 6 (Actual vs. Predicted plot) illustrates how the predictions of the best (Linear) and worst (MLP) models track the actual test data. The Linear model's predictions, while not perfect, follow the general trend of the actual prices much more closely than the MLP's, which show large deviations. 6. Discussion The results are striking and counter-intuitive. Despite the well-documented non-linearities in financial markets, the simplest model in our arsenal, Linear Regression, significantly outperformed all sophisticated non-linear and ensemble models. Several interpretations could explain this outcome as follows. i- Feature Restriction : The feature set was restricted to purely temporal elements (Year, Month, Day, Day-of-Week). The relationship between these features and the stock price may be predominantly linear for this particular index over the studied period. The complex models, designed to capture intricate non-linear patterns, likely overfit the training data in the absence of more predictive features (e.g., technical indicators, trading volumes, macroeconomic data). ii- Bias-Variance Trade-off : Linear models might have bias but low variance, which makes them robust to overfitting, especially on smaller or noisier datasets. The more complex models, with their high variance, may have learned the noise in the training period rather than the underlying signal, leading to poor generalization on the test set. iii: Default Parameters : The advanced models were used with their default parameters. It is possible that extensive hyperparameter tuning could have improved their performance and potentially closed the gap with the Linear Regression model. However, the fact that the default Linear model performed so well without any tuning is a result in itself. The key point to take away is a critical reminder for machine learning practice. Always begin with a simple baseline model. The assumption that a more complex algorithm will automatically yield better results is often incorrect. This baseline provides a crucial benchmark; any proposed complex model must convincingly outperform it to justify its additional complexity and computational cost. 7. Conclusion This study conducted a comparative analysis of eight machine learning models for forecasting the Dow Jones Australia Index. The findings clearly demonstrate that a simple Linear Regression model outperformed more complex algorithms, including XGBoost, Random Forest, and neural networks (MLP), when using only temporal features. This unexpected result highlights a fundamental principle in predictive modeling: complexity does not guarantee performance. The choice of model is profoundly influenced by the nature and predictive power of the feature set. In this case, the linear relationship between the engineered time features and the stock price was the most robust pattern for out-of-sample prediction. Future works might incorporate a wider range of predictors, such as lagged price values, moving averages, relative strength index (RSI), trading volume, and global economic indicators. Furthermore, a rigorous hyperparameter optimization routine should be applied to the non-linear models to ensure a fair comparison. Finally, exploring models specifically designed for time series, such as ARIMA, LSTMs, or Prophet, could provide a more suitable benchmark and potentially lead to improved forecasting accuracy. For predicting the DJ Australia index with basic temporal features, the most effective tool was also the simplest. This serves as a valuable lesson against the unnecessary complexity and underscores the enduring value of establishing a strong baseline. References Breiman, L. (2001), Random Forests, Machine Learning, Volume/Issue: Vol. 45, No. 1, Pages: 5–32, DOI: https://doi.org/10.1023/A:1010933404324, Link: https://link.springer.com/article/10.1023/A:1010933404324 Chen, T., & Guestrin, C. (2016), XGBoost: A Scalable Tree Boosting System, Conference: Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD '16), Pages: 785–794, DOI: https://doi.org/10.1145/2939672.2939785, Link: https://dl.acm.org/doi/10.1145/2939672.2939785 Dixon, M. F., Halperin, I., & Bilokon, P. (2020), Machine Learning in Finance, Springer, DOI: https://doi.org/10.1007/978-3-030-41068-1, Link: https://link.springer.com/book/10.1007/978-3-030-41068-1 Gandhmal, D. P., & Kumar, K. (2019), Systematic analysis and review of stock market prediction techniques, Computer Science Review, Volume/Issue: Vol. 34, Page: 100190, DOI: https://doi.org/10.1016/j.cosrev.2019.08.001, Link: https://www.sciencedirect.com/science/article/pii/S157401371930084X?via%3Dihub Hoseinzade, E., & Haratizadeh, S. (2019), CNNpred: CNN-based stock market prediction using a diverse set of variables, Expert Systems with Applications, Volume/Issue: Vol. 129, Pages: 273–285, DOI: https://doi.org/10.1016/j.eswa.2019.03.029, Link: https://www.sciencedirect.com/science/article/abs/pii/S0957417419301879 Investing.com (2025), Global Indices, accessed July 2, 2025, Link: https://www.investing.com/indices/global-indices Ke, G., Meng, Q., Finley, T., Wang, T., Chen, W., Ma, W., Ye, Q., & Liu, T.-Y. (2017), LightGBM: A Highly Efficient Gradient Boosting Decision Tree, Conference: Advances in Neural Information Processing Systems 30 (NIPS 2017), Pages: 3146–3154, Link: https://papers.nips.cc/paper/2017/hash/6449f44a102fde848669bdd9eb6b76fa-Abstract.html Kuskaya, S., Bilgili, F. (2025) Forecasting Electricity Price Index with Machine Learning Models and Strategies. Quality & Quantity, Springer, 2025, https://link.springer.com/journal/11135(forthcoming), PREPRINT (Version 1) available at Research Square, https://doi.org/10.21203/rs.3.rs-6298557/v1 Vashistha, D. (2025), Stock Price Prediction using Machine Learning in Python, Medium, Data Science Collective, May 6, 2025, Link: https://medium.com/data-science-collective/ Additional Declarations The authors declare no competing interests. Supplementary Files DowJonesAustraliaUSDHistoricalData.csv DataSource.docx Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {\"props\":{\"pageProps\":{\"initialData\":{\"identity\":\"rs-7473138\",\"acceptedTermsAndConditions\":true,\"allowDirectSubmit\":true,\"archivedVersions\":[],\"articleType\":\"Research Article\",\"associatedPublications\":[],\"authors\":[{\"id\":506471244,\"identity\":\"672ba019-e31d-48e2-b18a-7cf7e919f58a\",\"order_by\":0,\"name\":\"Sevda Kuşkaya\",\"email\":\"\",\"orcid\":\"https://orcid.org/0000-0003-4527-5713\",\"institution\":\"Erciyes University\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Sevda\",\"middleName\":\"\",\"lastName\":\"Kuşkaya\",\"suffix\":\"\"},{\"id\":506471245,\"identity\":\"fa7341a5-81c3-4c7c-a971-3c7f8e1eb8ed\",\"order_by\":1,\"name\":\"Faik Bilgili\",\"email\":\"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA/0lEQVRIie3QMYvCMBTA8ReEdgl0fYeF+wqFgjgU81USDjI5HAjOOYR2cr/Nz+Di3FLQRW7udNDFyaGCQwfBSzg8cWi6Cpf/9Bryo0kAXK4nzP8kHzlEYxqYL/67GFkJrYjSBOmLeiDcRgByAIQovy/2kOFCFed3DOPdsm7qNGHgLzZI2u9uEhaqDPXBRvtdjCKVQtHtHIHPOglDoUo0pJIeiLTkgNORJt0nozcSrw6GXBm8HvtJ0WgSoWdITpT+o52Yu5hHxr0cIP96EymVszGXFjLM6rq9JCzItuTUzics8Mt11STdxDSgt4l4AJ4Z7EBvbP/GS89Wl8vl+pf9AAmrTazFi88MAAAAAElFTkSuQmCC\",\"orcid\":\"https://orcid.org/0000-0003-4138-6897\",\"institution\":\"Erciyes University\",\"correspondingAuthor\":true,\"prefix\":\"\",\"firstName\":\"Faik\",\"middleName\":\"\",\"lastName\":\"Bilgili\",\"suffix\":\"\"}],\"badges\":[],\"createdAt\":\"2025-08-27 15:32:34\",\"currentVersionCode\":1,\"declarations\":{\"humanSubjects\":false,\"vertebrateSubjects\":false,\"conflictsOfInterestStatement\":false,\"humanSubjectEthicalGuidelines\":false,\"humanSubjectConsent\":false,\"humanSubjectClinicalTrial\":false,\"humanSubjectCaseReport\":false,\"vertebrateSubjectEthicalGuidelines\":false},\"doi\":\"10.21203/rs.3.rs-7473138/v1\",\"doiUrl\":\"https://doi.org/10.21203/rs.3.rs-7473138/v1\",\"draftVersion\":[],\"editorialEvents\":[],\"editorialNote\":\"\",\"failedWorkflow\":false,\"files\":[{\"id\":90083488,\"identity\":\"3d2a562d-6e42-4a22-bc15-96cefe24215d\",\"added_by\":\"auto\",\"created_at\":\"2025-08-28 09:28:23\",\"extension\":\"png\",\"order_by\":1,\"title\":\"Figure 1\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":72349,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003e\\u003cstrong\\u003eDow Jones Australia Index (USD) Historical Price Trend\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eData Source: Investing.com Global Indices: DJ Australia\\u003c/p\\u003e\\n\\u003cp\\u003ehttps://www.investing.com/indices/dj-australia\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"1.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7473138/v1/98f175f6deb5c793c8fbd5e7.png\"},{\"id\":90083841,\"identity\":\"047b3823-be40-4735-8801-eee70f659967\",\"added_by\":\"auto\",\"created_at\":\"2025-08-28 09:36:23\",\"extension\":\"png\",\"order_by\":2,\"title\":\"Figure 2\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":57806,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003e\\u003cstrong\\u003eDistribution Plots of Financial Features\\u003c/strong\\u003e\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"2.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7473138/v1/be64545306e84d077894d448.png\"},{\"id\":90083840,\"identity\":\"29872500-496a-46f5-b0c1-55421bb0ef7e\",\"added_by\":\"auto\",\"created_at\":\"2025-08-28 09:36:23\",\"extension\":\"png\",\"order_by\":3,\"title\":\"Figure 3\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":24366,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003e\\u003cstrong\\u003eBox Plots of Financial Features\\u003c/strong\\u003e\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"3.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7473138/v1/aa1cc02317915bcba15b53bb.png\"},{\"id\":90083494,\"identity\":\"098e66b3-739b-4b0f-94ae-2c42847f613c\",\"added_by\":\"auto\",\"created_at\":\"2025-08-28 09:28:23\",\"extension\":\"png\",\"order_by\":4,\"title\":\"Figure 4\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":24686,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003e\\u003cstrong\\u003eModel Forecast Accuracy Comparison (RMSE, MAE, MAPE)\\u003c/strong\\u003e\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"4.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7473138/v1/1f39133942a51537724087ec.png\"},{\"id\":90084654,\"identity\":\"b6c57662-bee4-4f46-98f8-41a6e18bc0ee\",\"added_by\":\"auto\",\"created_at\":\"2025-08-28 09:44:23\",\"extension\":\"png\",\"order_by\":5,\"title\":\"Figure 5\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":18894,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003e\\u003cstrong\\u003eModel Prediction Accuracy Comparison (MAPE)\\u003c/strong\\u003e\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"5.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7473138/v1/0a83add6fe5eaef6718d8b16.png\"},{\"id\":90083849,\"identity\":\"1593ae78-c0e4-4b08-9810-e3072f4ffcd7\",\"added_by\":\"auto\",\"created_at\":\"2025-08-28 09:36:24\",\"extension\":\"png\",\"order_by\":6,\"title\":\"Figure 6\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":61204,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003e\\u003cstrong\\u003eActual vs. Predicted Stock Prices for Time Index Test Set (Actual, Linear prediction, MLP prediction)\\u003c/strong\\u003e\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"6.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7473138/v1/95865792a7f9240bc63cab5b.png\"},{\"id\":90084850,\"identity\":\"052f4d6b-5781-4920-bc9e-af8d03104d21\",\"added_by\":\"auto\",\"created_at\":\"2025-08-28 09:52:24\",\"extension\":\"pdf\",\"order_by\":0,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"manuscript-pdf\",\"size\":903430,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"manuscript.pdf\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7473138/v1/40db5ee4-a7a0-4160-be1f-9abc36d7f9e5.pdf\"},{\"id\":90083501,\"identity\":\"97d3a9cd-cb78-490a-a022-52bd5160f515\",\"added_by\":\"auto\",\"created_at\":\"2025-08-28 09:28:23\",\"extension\":\"csv\",\"order_by\":1,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"supplement\",\"size\":159126,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"DowJonesAustraliaUSDHistoricalData.csv\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7473138/v1/6171cb7519789f6517ae9c3b.csv\"},{\"id\":90083491,\"identity\":\"27bda49d-2919-4f11-8268-e16f6f08ab8a\",\"added_by\":\"auto\",\"created_at\":\"2025-08-28 09:28:23\",\"extension\":\"docx\",\"order_by\":2,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"supplement\",\"size\":14499,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"DataSource.docx\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7473138/v1/079270e2c2392c7485d3682b.docx\"}],\"financialInterests\":\"The authors declare no competing interests.\",\"formattedTitle\":\"\\u003cp\\u003e\\u003cstrong\\u003eForecasting the Dow Jones Australia Index: A Comparative Evaluation of Machine Learning Regression Models\\u003c/strong\\u003e\\u003c/p\\u003e\",\"fulltext\":[{\"header\":\"1. Introduction\",\"content\":\"\\u003cp\\u003eThe inherent volatility and complex, non-linear nature of stock markets make their prediction a challenging yet highly sought-after goal for investors, analysts, and economists. The ability to forecast market movements, even with marginal accuracy, can provide a significant advantage. The Dow Jones Australia Index, a key indicator of the Australian equity market's health, is influenced by a multitude of factors, including global economic trends, commodity prices, and domestic policy changes.\\u003c/p\\u003e\\u003cp\\u003eWith the advent of powerful computational resources and sophisticated algorithms, machine learning (ML) has emerged as a promising tool for uncovering hidden patterns in financial data that traditional statistical methods might miss. Models like Random Forests, Gradient Boosting Machines (e.g., XGBoost, LightGBM, and CatBoost), and neural networks are increasingly being applied to forecast financial time series.\\u003c/p\\u003e\\u003cp\\u003eThis study aims to contribute to this field by conducting a comprehensive empirical evaluation of multiple machine learning regression models for predicting the daily closing price of the DJ Australia index. The primary objective is to identify the model that provides the most accurate forecasts based on a set of engineered temporal features. The surprising outcome, which saw a simple Linear Regression model outperform more complex alternatives, provides a critical insight into the nature of this particular forecasting problem and highlights a crucial lesson in model selection.\\u003c/p\\u003e\"},{\"header\":\"2. Literature Review\",\"content\":\"\\u003cp\\u003eThe application of machine learning to stock price prediction is a well-established domain within computational finance. Early work often focused on traditional time series models like ARIMA (Auto Regressive Integrated Moving Average). However, their limitations in capturing non-linear relationships and complex market dynamics led researchers to explore ML techniques\\u003c/p\\u003e\\u003cp\\u003eNumerous studies have investigated the efficacy of various models. Support Vector Machines (SVM), particularly SVR for regression tasks, have been used for forecasting due to their effectiveness in high-dimensional spaces (Gandhmal \\u0026amp; Kumar, \\u003cspan citationid=\\\"CR7\\\" class=\\\"CitationRef\\\"\\u003e2019\\u003c/span\\u003e). Recently, ensemble methods have gained prominence. Random Forests, introduced by Breiman (2001), are praised for their robustness against overfitting. Gradient boosting frameworks like XGBoost (Chen \\u0026amp; Guestrin, \\u003cspan citationid=\\\"CR4\\\" class=\\\"CitationRef\\\"\\u003e2016\\u003c/span\\u003e) and LightGBM (Ke et al., \\u003cspan citationid=\\\"CR14\\\" class=\\\"CitationRef\\\"\\u003e2017\\u003c/span\\u003e) have consistently delivered state-of-the-art results in various Kaggle competitions, including those involving financial data, due to their speed and performance.\\u003c/p\\u003e\\u003cp\\u003eDeep learning approaches, such as Multi-layer Perceptrons (MLP) and Recurrent Neural Networks (RNNs), have also been applied to model temporal dependencies in sequential data like stock prices (Hoseinzade \\u0026amp; Haratizadeh, \\u003cspan citationid=\\\"CR10\\\" class=\\\"CitationRef\\\"\\u003e2019\\u003c/span\\u003e). Despite the sophistication of these models, the literature also documents instances where simpler models can perform on par with or even outperform complex ones, especially when the feature set is limited or the underlying relationship is linear. Dixon et al. (\\u003cspan citationid=\\\"CR6\\\" class=\\\"CitationRef\\\"\\u003e2020\\u003c/span\\u003e) provide a broad overview of ML in finance and discuss the practical considerations and pitfalls, such as the fact that simpler models can be more robust and effective. This body of work validates a comparative approach but also cautions against assuming that superior complexity always yields superior results. Vashistha (\\u003cspan citationid=\\\"CR16\\\" class=\\\"CitationRef\\\"\\u003e2025\\u003c/span\\u003e) underlines that ML can predict whether a particular stock trade will be profitable or not and shows how to predict a signal of stock prices.\\u003c/p\\u003e\\u003cp\\u003eKuşkaya and Bilgili (\\u003cspan citationid=\\\"CR15\\\" class=\\\"CitationRef\\\"\\u003e2025\\u003c/span\\u003e) found that among 19 ML models, the Linear Regression and CatBoost Regressor models achieved the best results in terms of accuracy and computational efficiency. Additionally, they noted that the Linear Regression model using the TimeSeriesSplit strategy demonstrated high ex-post and ex-ante forecast accuracy for the US electricity prices index.\\u003c/p\\u003e\\u003cp\\u003eOur study adds to this literature by providing a direct, head-to-head comparison of these diverse model families on the DJ Australia index, using a minimalist feature set based solely on time, and by explicitly documenting a case where parsimony prevails.\\u003c/p\\u003e\"},{\"header\":\"3. Data\",\"content\":\"\\u003cp\\u003eThe data for this study were sourced from a CSV/Excel file of DJ Australia (Investing.com) containing daily historical data for the Dow Jones Australia Index (USD). The original dataset comprised 2,590 entries (rows) spanning from July 28, 2015, to June 30, 2025. Each entry included the following financial variables:\\u003c/p\\u003e\\u003cp\\u003eDate: The trading day (recorded as an object and later converted to datetime).\\u003c/p\\u003e\\u003cp\\u003ePrice: The daily closing price of the index (the target variable for prediction).\\u003c/p\\u003e\\u003cp\\u003eOpen: The daily opening price.\\u003c/p\\u003e\\u003cp\\u003eHigh: The highest price reached during the day.\\u003c/p\\u003e\\u003cp\\u003eLow: The lowest price reached during the day.\\u003c/p\\u003e\\u003cp\\u003eChange %: The percentage change from the previous day's closing price.\\u003c/p\\u003e\\u003cp\\u003eA preliminary exploratory data analysis (EDA) was conducted. The data was checked for missing values, and one null entry in the 'Price' column was identified and subsequently removed to ensure data integrity. Distribution plots (histograms) and boxplots were generated for all numerical variables to understand their spread and identify potential outliers. The initial time series plot of the closing price (Fig.\\u0026nbsp;1) confirmed a clear upward trend over the entire period, which is a common characteristic of major equity indices over the long term.\\u003c/p\\u003e\\u003cp\\u003eFigure 1 presents the daily closing price (Price) of the Dow Jones Australia Index from July 2015 to June 2025. The x-axis represents time (Date), and the y-axis represents the price in USD. The plot reveals a strong, persistent upward trend over the entire decade, which is a common characteristic of major equity indices over the long term, reflecting economic growth and market expansion. Except for one business day observation on March 23, 2020 (pandemic crash), the graph shows a relatively steady growth period for the Australian market during this time. This visual establishes the non-stationary nature of the data, a key challenge in time series forecasting.\\u003c/p\\u003e\\u003cp\\u003eThis panel of histograms shows the frequency distribution of the key financial variables in the dataset: Price, Open, High, Low, and Change %. The most critical observation is the bimodal distribution (two distinct peaks) for all four variables. This strongly suggests that the market exhibited two different \\\"regimes\\\" or phases during the 2015\\u0026ndash;2025 period:\\u003c/p\\u003e\\u003cp\\u003eThe first peak of the price regime (lower price regime) likely corresponds to the earlier years in the dataset, such as 2015\\u0026ndash;2019.\\u003c/p\\u003e\\u003cp\\u003eA higher-price regime (second peak) likely refers to the later years (e.g., post-2020 recovery and subsequent growth).\\u003c/p\\u003e\\u003cp\\u003eThis multimodality indicates a significant structural shift in the market, potentially driven by major economic events (e.g., the post-pandemic recovery). This insight is crucial because it challenges the assumption of a single, stable distribution in many forecasting models.\\u003c/p\\u003e\\u003cp\\u003eFigure 3 of Box Plots of Financial Features visualizes the spread and central tendency of the same financial features and identifies potential outliers. The box represents the interquartile range (IQR-the middle 50% of the data), the line inside the box is the median, and the \\u0026lsquo;whiskers\\u0026rsquo; extend to show the range of typical data points. Points beyond the whiskers are considered outliers.\\u003c/p\\u003e\\u003cp\\u003eGiven the strong correlation between these values on any given day, the Price, Open, High, and Low plots are nearly identical. They show a positive skew (the median is closer to the bottom of the box, and the upper whisker is longer), consistent with the upward trend seen in Fig.\\u0026nbsp;1. There are a few extreme outliers.\\u003c/p\\u003e\\u003cp\\u003e\\u003c/p\\u003e\\u003cp\\u003eThe change % plot is centered around zero, with a symmetrical distribution. The presence of numerous outliers on both the positive and negative sides is normal for daily percentage change data, reflecting days of unusually large gains or losses.\\u003c/p\\u003e\"},{\"header\":\"4. Methodology\",\"content\":\"\\u003cp\\u003eThe overall methodology followed a standard supervised machine learning workflow for regression, tailored for time-series data to avoid look-ahead bias.\\u003c/p\\u003e\\u003cdiv id=\\\"Sec5\\\" class=\\\"Section2\\\"\\u003e\\u003ch2\\u003e4.1. Data Preprocessing and Feature Engineering:\\u003c/h2\\u003e\\u003cp\\u003eThe 'Date' column was converted into a datetime object and used as the basis for sorting the entire dataset chronologically in ascending order. This is a critical step for time-series analysis. The index was reset to ensure proper alignment for subsequent operations and plotting.\\u003c/p\\u003e\\u003cp\\u003eInstead of using the raw financial features (Open, High, Low, Change %), which were discarded to avoid leakage and simplify the initial model comparison, the Date was decomposed into numerical features. This feature engineering step extracted the following explanatory variables: Year, Month, Day, and Day-of-Week (Monday\\u0026thinsp;=\\u0026thinsp;0, Sunday\\u0026thinsp;=\\u0026thinsp;6). This transformed the problem into predicting the index price based solely on temporal components.\\u003c/p\\u003e\\u003c/div\\u003e\\u003cdiv id=\\\"Sec6\\\" class=\\\"Section2\\\"\\u003e\\u003ch2\\u003e4.2. Train-Test Split:\\u003c/h2\\u003e\\u003cp\\u003eTo preserve the temporal order of the data and simulate a realistic forecasting scenario, a sequential split was used instead of a random shuffle. The first 80% of the data (2015\\u0026ndash;2023) was used for training, and the last 20% (2023\\u0026ndash;2025) was used for testing. This ensures that models are trained on past data and evaluated on future, unseen data.\\u003c/p\\u003e\\u003c/div\\u003e\\u003cdiv id=\\\"Sec7\\\" class=\\\"Section2\\\"\\u003e\\u003ch2\\u003e\\u003cb\\u003e4.3.\\u003c/b\\u003e Model Selection and Training:\\u003c/h2\\u003e\\u003cp\\u003eA diverse set of eight regression algorithms was selected to represent different modeling paradigms:\\u003c/p\\u003e\\u003col style=\\\"list-style-type: lower-roman;\\\"\\u003e\\n \\u003cli\\u003e\\u003cstrong\\u003eLinear Models\\u003c/strong\\u003e: Linear Regression\\u003c/li\\u003e\\n \\u003cli\\u003e\\u003cstrong\\u003eSupport Vector Machines\\u003c/strong\\u003e: Support Vector Regression (SVR)\\u003c/li\\u003e\\n \\u003cli\\u003e\\u003cstrong\\u003eEnsemble Methods\\u003c/strong\\u003e: Random Forest (RF), XGBoost (XGB), LightGBM, CatBoost\\u003c/li\\u003e\\n \\u003cli\\u003e\\u003cstrong\\u003eInstance-Based Learning\\u003c/strong\\u003e: k-Nearest Neighbors (KNN)\\u003c/li\\u003e\\n \\u003cli\\u003e\\u003cstrong\\u003eNeural Networks\\u003c/strong\\u003e: Multi-layer Perceptron (MLP)\\u003c/li\\u003e\\n\\u003c/ol\\u003e\\u003cp\\u003eAll models were trained on the same feature set (Year, Month, Day, Day-of-Week) and target variable (Price) using their default hyperparameters in Scikit-Learn to provide a baseline comparison of their out-of-the-box performance.\\u003c/p\\u003e\\u003c/div\\u003e\\u003cdiv id=\\\"Sec8\\\" class=\\\"Section2\\\"\\u003e\\u003ch2\\u003e4.4. Performance Evaluation:\\u003c/h2\\u003e\\u003cp\\u003eModel performance was evaluated on the test set using a comprehensive set of metrics to assess different aspects of prediction error:\\u003c/p\\u003e\\u003cp\\u003e\\u003cul\\u003e\\u003cli\\u003e\\u003cp\\u003eRoot Mean Squared Error \\u003cb\\u003e(RMSE)\\u003c/b\\u003e: It prioritizes larger errors, useful for understanding typical error magnitude in the units of the target variable.\\u003c/p\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cp\\u003eMean Absolute Error \\u003cb\\u003e(MAE)\\u003c/b\\u003e: It provides a linear score for the average error magnitude.\\u003c/p\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cp\\u003eMean Absolute Percentage Error \\u003cb\\u003e(MAPE)\\u003c/b\\u003e: It expresses error as a percentage, making it easily interpretable.\\u003c/p\\u003e\\u003c/li\\u003e\\u003c/ul\\u003e\\u003c/p\\u003e\\u003cp\\u003eThe models were ranked primarily by RMSE for final comparison. The results were compiled into a comparative table and visualized using bar charts for error metrics and a time series plot comparing the actual values to the predictions of the best- and worst-performing models.\\u003c/p\\u003e\\u003cp\\u003e\\u003cdiv class=\\\"gridtable\\\"\\u003e\\u003ctable float=\\\"Yes\\\" id=\\\"Tab1\\\" border=\\\"1\\\"\\u003e\\u003ccaption language=\\\"En\\\"\\u003e\\u003cdiv class=\\\"CaptionNumber\\\"\\u003eTable 1\\u003c/div\\u003e\\u003cdiv class=\\\"CaptionContent\\\"\\u003e\\u003cp\\u003eML Model Comparison Results\\u003c/p\\u003e\\u003c/div\\u003e\\u003c/caption\\u003e\\u003ccolgroup cols=\\\"5\\\"\\u003e\\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c1\\\" colnum=\\\"1\\\"\\u003e\\u003c/div\\u003e\\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c2\\\" colnum=\\\"2\\\"\\u003e\\u003c/div\\u003e\\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c3\\\" colnum=\\\"3\\\"\\u003e\\u003c/div\\u003e\\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c4\\\" colnum=\\\"4\\\"\\u003e\\u003c/div\\u003e\\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c5\\\" colnum=\\\"5\\\"\\u003e\\u003c/div\\u003e\\u003cthead\\u003e\\u003ctr\\u003e\\u003cth align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u0026nbsp;\\u003c/th\\u003e\\u003cth align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eModel\\u003c/p\\u003e\\u003c/th\\u003e\\u003cth align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003eRMSE\\u003c/p\\u003e\\u003c/th\\u003e\\u003cth align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003eMAE\\u003c/p\\u003e\\u003c/th\\u003e\\u003cth align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003eMAPE\\u003c/p\\u003e\\u003c/th\\u003e\\u003c/tr\\u003e\\u003c/thead\\u003e\\u003ctbody\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003e1\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eLinear\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e29.853786\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003e23.2086\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e7.054014\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003e2\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eKNN\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e58.521712\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003e50.953445\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e15.257116\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003e3\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eCatBoost\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e58.554258\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003e50.946156\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e15.254353\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003e4\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eXGB\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e59.847303\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003e52.40042\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e15.665139\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003e5\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eRF\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e59.886671\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003e52.149562\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e15.605581\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003e6\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eLightGBM\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e60.150836\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003e53.04236\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e15.843552\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003e7\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eSVR\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e83.57932\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003e78.942095\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e23.261142\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003e8\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eMLP\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e85.142504\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003e80.620807\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e23.739671\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003c/tbody\\u003e\\u003c/colgroup\\u003e\\u003c/table\\u003e\\u003c/div\\u003e\\u003c/p\\u003e\\u003cp\\u003eIn Figs.\\u0026nbsp;4 and 5, the bar charts provide a visual comparison of the forecasting errors for the eight tested models, ranked from best (Linear) to worst (MLP).\\u003c/p\\u003e\\u003cp\\u003eFigure 4 compares three error metrics (forecast/prediction accuracy metrics) simultaneously: RMSE, MAE, and MAPE. The dramatic difference in the bar for the Linear Regression model across all three metrics immediately conveys its superior performance.\\u003c/p\\u003e\\u003cp\\u003eFigure 5 reveals the MAPE (Mean Absolute Percentage Error). This is often the most intuitive metric, as it expresses the error as a percentage. It clearly shows that Linear Regression's average error is around 7%, while all other models have an error of approximately 15% or higher. This chart effectively communicates that the simple linear model is, on average, more than twice as accurate as the complex models for this specific task.\\u003c/p\\u003e\\u003cp\\u003e\\u003c/p\\u003e\\u003cp\\u003e\\u003c/p\\u003e\\u003cp\\u003eIn Fig.\\u0026nbsp;6, the Actual vs. Predicted Prices time series plot is perhaps the most intuitive illustration of model performance. It shows the actual values of the index in the test set (the holdout period from 2023 to 2025) compared to the predictions made by the best model (Linear Regression) and the worst model (Multi-layer Perceptron - MLP). The black line (Actual) shows the true market movement. The dashed line (Linear Prediction) closely follows the black line's general direction and turning points. While it doesn't capture every minor fluctuation, it accurately tracks the overall trend and rhythm of the market.\\u003c/p\\u003e\\u003cp\\u003eThe dotted line (MLP Prediction) is highly erratic and fails to follow the actual data meaningfully. It demonstrates severe overfitting; the complex MLP model learned noise from the training data instead of the underlying pattern, making its predictions on new, unseen data (the test set) essentially useless.\\u003c/p\\u003e\\u003cp\\u003eThis visual starkly contrasts a model that generalized well (Linear) with one that failed (MLP), powerfully reinforcing the paper's central finding\\u003c/p\\u003e\\u003cp\\u003e\\u003c/p\\u003e\\u003c/div\\u003e\"},{\"header\":\"5. Results\",\"content\":\"\\u003cp\\u003eThe dataset consisted of 2,590 daily records from July 2015 to June 2025. After sorting by date and splitting sequentially (80% train, 20% test), the training set contained 2,071 samples and the test set 519 samples. Eight regression models were trained and evaluated on the test set.\\u003c/p\\u003e\\u003cp\\u003eThe performance metrics for all models are summarized in Table\\u0026nbsp;1. The models are ranked from best to worst based on RMSE.\\u003c/p\\u003e\\u003cp\\u003e\\u003cstrong\\u003eBest Performing Model\\u003c/strong\\u003e\\u003cp\\u003eLinear Regression was the clear top performer. It achieved an RMSE of 29.853, an MAE of 23.208, and a MAPE of 7.05%.\\u003c/p\\u003e\\u003c/p\\u003e\\u003cp\\u003e\\u003cstrong\\u003eIntermediate Performers\\u003c/strong\\u003e\\u003cp\\u003eKNN, CatBoost, XGBoost, and Random Forest formed a middle group. Their error metrics were roughly twice as high as those of Linear Regression (RMSE\\u0026thinsp;~\\u0026thinsp;58.52\\u0026ndash;59.89, MAPE\\u0026thinsp;~\\u0026thinsp;15.26\\u0026ndash;15.60%).\\u003c/p\\u003e\\u003c/p\\u003e\\u003cp\\u003e\\u003cstrong\\u003eWorst Performing Models\\u003c/strong\\u003e\\u003cp\\u003eSVR and MLP were the least accurate models, with the highest errors (RMSE\\u0026thinsp;\\u0026gt;\\u0026thinsp;83, MAPE\\u0026thinsp;\\u0026gt;\\u0026thinsp;23%), indicating they performed worse than a simple horizontal line predicting the mean.\\u003c/p\\u003e\\u003c/p\\u003e\\u003cp\\u003eFigures 4 and 5 (metrics comparison bar charts) visually confirm the significant advantage of the Linear model. Figure\\u0026nbsp;6 (Actual vs. Predicted plot) illustrates how the predictions of the best (Linear) and worst (MLP) models track the actual test data. The Linear model's predictions, while not perfect, follow the general trend of the actual prices much more closely than the MLP's, which show large deviations.\\u003c/p\\u003e\"},{\"header\":\"6. Discussion\",\"content\":\"\\u003cp\\u003eThe results are striking and counter-intuitive. Despite the well-documented non-linearities in financial markets, the simplest model in our arsenal, Linear Regression, significantly outperformed all sophisticated non-linear and ensemble models.\\u003c/p\\u003e\\u003cp\\u003eSeveral interpretations could explain this outcome as follows.\\u003c/p\\u003e\\u003cp\\u003ei- \\u003cb\\u003eFeature Restriction\\u003c/b\\u003e: The feature set was restricted to purely temporal elements (Year, Month, Day, Day-of-Week). The relationship between these features and the stock price may be predominantly linear for this particular index over the studied period. The complex models, designed to capture intricate non-linear patterns, likely overfit the training data in the absence of more predictive features (e.g., technical indicators, trading volumes, macroeconomic data).\\u003c/p\\u003e\\u003cp\\u003eii- \\u003cb\\u003eBias-Variance Trade-off\\u003c/b\\u003e: Linear models might have bias but low variance, which makes them robust to overfitting, especially on smaller or noisier datasets. The more complex models, with their high variance, may have learned the noise in the training period rather than the underlying signal, leading to poor generalization on the test set.\\u003c/p\\u003e\\u003cp\\u003eiii: \\u003cb\\u003eDefault Parameters\\u003c/b\\u003e: The advanced models were used with their default parameters. It is possible that extensive hyperparameter tuning could have improved their performance and potentially closed the gap with the Linear Regression model. However, the fact that the default Linear model performed so well without any tuning is a result in itself.\\u003c/p\\u003e\\u003cp\\u003eThe key point to take away is a critical reminder for machine learning practice. Always begin with a simple baseline model. The assumption that a more complex algorithm will automatically yield better results is often incorrect. This baseline provides a crucial benchmark; any proposed complex model must convincingly outperform it to justify its additional complexity and computational cost.\\u003c/p\\u003e\"},{\"header\":\"7. Conclusion\",\"content\":\"\\u003cp\\u003eThis study conducted a comparative analysis of eight machine learning models for forecasting the Dow Jones Australia Index. The findings clearly demonstrate that a simple Linear Regression model outperformed more complex algorithms, including XGBoost, Random Forest, and neural networks (MLP), when using only temporal features.\\u003c/p\\u003e\\u003cp\\u003eThis unexpected result highlights a fundamental principle in predictive modeling: complexity does not guarantee performance. The choice of model is profoundly influenced by the nature and predictive power of the feature set. In this case, the linear relationship between the engineered time features and the stock price was the most robust pattern for out-of-sample prediction.\\u003c/p\\u003e\\u003cp\\u003eFuture works might incorporate a wider range of predictors, such as lagged price values, moving averages, relative strength index (RSI), trading volume, and global economic indicators. Furthermore, a rigorous hyperparameter optimization routine should be applied to the non-linear models to ensure a fair comparison. Finally, exploring models specifically designed for time series, such as ARIMA, LSTMs, or Prophet, could provide a more suitable benchmark and potentially lead to improved forecasting accuracy.\\u003c/p\\u003e\\u003cp\\u003eFor predicting the DJ Australia index with basic temporal features, the most effective tool was also the simplest. This serves as a valuable lesson against the unnecessary complexity and underscores the enduring value of establishing a strong baseline.\\u003c/p\\u003e\"},{\"header\":\"References\",\"content\":\"\\u003col\\u003e\\n\\u003cli\\u003eBreiman, L. (2001), Random Forests, Machine Learning, Volume/Issue: Vol. 45, No. 1, Pages: 5\\u0026ndash;32, DOI: https://doi.org/10.1023/A:1010933404324, Link: https://link.springer.com/article/10.1023/A:1010933404324\\u003c/li\\u003e\\n\\u003cli\\u003eChen, T., \\u0026amp; Guestrin, C. (2016), XGBoost: A Scalable Tree Boosting System, Conference: Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD \\u0026apos;16), Pages: 785\\u0026ndash;794, DOI: https://doi.org/10.1145/2939672.2939785, Link: https://dl.acm.org/doi/10.1145/2939672.2939785\\u003c/li\\u003e\\n\\u003cli\\u003eDixon, M. F., Halperin, I., \\u0026amp; Bilokon, P. (2020), Machine Learning in Finance, Springer, DOI: https://doi.org/10.1007/978-3-030-41068-1, Link: https://link.springer.com/book/10.1007/978-3-030-41068-1\\u003c/li\\u003e\\n\\u003cli\\u003eGandhmal, D. P., \\u0026amp; Kumar, K. (2019), Systematic analysis and review of stock market prediction techniques, Computer Science Review, Volume/Issue: Vol. 34, Page: 100190, DOI: https://doi.org/10.1016/j.cosrev.2019.08.001, Link: https://www.sciencedirect.com/science/article/pii/S157401371930084X?via%3Dihub\\u003c/li\\u003e\\n\\u003cli\\u003eHoseinzade, E., \\u0026amp; Haratizadeh, S. (2019), CNNpred: CNN-based stock market prediction using a diverse set of variables, Expert Systems with Applications, Volume/Issue: Vol. 129, Pages: 273\\u0026ndash;285, DOI: https://doi.org/10.1016/j.eswa.2019.03.029, Link: https://www.sciencedirect.com/science/article/abs/pii/S0957417419301879\\u003c/li\\u003e\\n\\u003cli\\u003eInvesting.com (2025), Global Indices, accessed July 2, 2025, Link: https://www.investing.com/indices/global-indices\\u003c/li\\u003e\\n\\u003cli\\u003eKe, G., Meng, Q., Finley, T., Wang, T., Chen, W., Ma, W., Ye, Q., \\u0026amp; Liu, T.-Y. (2017), LightGBM: A Highly Efficient Gradient Boosting Decision Tree, Conference: Advances in Neural Information Processing Systems 30 (NIPS 2017), Pages: 3146\\u0026ndash;3154, Link: https://papers.nips.cc/paper/2017/hash/6449f44a102fde848669bdd9eb6b76fa-Abstract.html\\u003c/li\\u003e\\n\\u003cli\\u003eKuskaya, S., Bilgili, F. (2025) Forecasting Electricity Price Index with Machine Learning Models and Strategies. Quality \\u0026amp; Quantity, Springer, 2025, https://link.springer.com/journal/11135(forthcoming), PREPRINT (Version 1) available at Research Square, https://doi.org/10.21203/rs.3.rs-6298557/v1\\u003c/li\\u003e\\n\\u003cli\\u003eVashistha, D. (2025), Stock Price Prediction using Machine Learning in Python, Medium, Data Science Collective, May 6, 2025, Link: https://medium.com/data-science-collective/\\u003c/li\\u003e\\n\\u003c/ol\\u003e\"}],\"fulltextSource\":\"\",\"fullText\":\"\",\"funders\":[],\"hasAdminPriorityOnWorkflow\":false,\"hasManuscriptDocX\":true,\"hasOptedInToPreprint\":true,\"hasPassedJournalQc\":\"\",\"hasAnyPriority\":true,\"hideJournal\":true,\"highlight\":\"\",\"institution\":\"\",\"isAcceptedByJournal\":false,\"isAuthorSuppliedPdf\":false,\"isDeskRejected\":\"\",\"isHiddenFromSearch\":false,\"isInQc\":false,\"isInWorkflow\":true,\"isPdf\":false,\"isPdfUpToDate\":true,\"isWithdrawnOrRetracted\":false,\"journal\":{\"display\":true,\"email\":\"info@researchsquare.com\",\"identity\":\"researchsquare\",\"isNatureJournal\":false,\"hasQc\":true,\"allowDirectSubmit\":true,\"externalIdentity\":\"\",\"sideBox\":\"\",\"snPcode\":\"\",\"submissionUrl\":\"/submission\",\"title\":\"Research Square\",\"twitterHandle\":\"researchsquare\",\"acdcEnabled\":true,\"dfaEnabled\":false,\"editorialSystem\":\"\",\"reportingPortfolio\":\"\",\"inReviewEnabled\":false,\"inReviewRevisionsEnabled\":true},\"keywords\":\"Stock Market Prediction, Machine Learning, Time Series Forecasting, Linear Regression, Model Comparison, Dow Jones Australia Index\",\"lastPublishedDoi\":\"10.21203/rs.3.rs-7473138/v1\",\"lastPublishedDoiUrl\":\"https://doi.org/10.21203/rs.3.rs-7473138/v1\",\"license\":{\"name\":\"CC BY 4.0\",\"url\":\"https://creativecommons.org/licenses/by/4.0/\"},\"manuscriptAbstract\":\"\\u003cp\\u003eAn 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.\\u003c/p\\u003e\",\"manuscriptTitle\":\"Forecasting the Dow Jones Australia Index: A Comparative Evaluation of Machine Learning Regression Models\",\"msid\":\"\",\"msnumber\":\"\",\"nonDraftVersions\":[{\"code\":1,\"date\":\"2025-08-28 09:28:19\",\"doi\":\"10.21203/rs.3.rs-7473138/v1\",\"editorialEvents\":[{\"type\":\"communityComments\",\"content\":0}],\"status\":\"published\",\"journal\":{\"display\":true,\"email\":\"info@researchsquare.com\",\"identity\":\"researchsquare\",\"isNatureJournal\":false,\"hasQc\":true,\"allowDirectSubmit\":true,\"externalIdentity\":\"\",\"sideBox\":\"\",\"snPcode\":\"\",\"submissionUrl\":\"/submission\",\"title\":\"Research Square\",\"twitterHandle\":\"researchsquare\",\"acdcEnabled\":true,\"dfaEnabled\":false,\"editorialSystem\":\"\",\"reportingPortfolio\":\"\",\"inReviewEnabled\":false,\"inReviewRevisionsEnabled\":true}}],\"origin\":\"\",\"ownerIdentity\":\"5626b3cb-198f-4912-a58e-7ee332108360\",\"owner\":[],\"postedDate\":\"August 28th, 2025\",\"published\":true,\"recentEditorialEvents\":[],\"rejectedJournal\":[],\"revision\":\"\",\"amendment\":\"\",\"status\":\"posted\",\"subjectAreas\":[{\"id\":53798109,\"name\":\"Artificial Intelligence and Machine Learning\"},{\"id\":53798110,\"name\":\"Econometrics\"},{\"id\":53798111,\"name\":\"Macroeconomics\"},{\"id\":53798112,\"name\":\"Applied Statistics\"},{\"id\":53798113,\"name\":\"Finance\"},{\"id\":53798114,\"name\":\"Computational Mathematics\"}],\"tags\":[],\"updatedAt\":\"2025-08-28T09:28:19+00:00\",\"versionOfRecord\":[],\"versionCreatedAt\":\"2025-08-28 09:28:19\",\"video\":\"\",\"vorDoi\":\"\",\"vorDoiUrl\":\"\",\"workflowStages\":[]},\"version\":\"v1\",\"identity\":\"rs-7473138\",\"journalConfig\":\"researchsquare\"},\"__N_SSP\":true},\"page\":\"/article/[identity]/[[...version]]\",\"query\":{\"redirect\":\"/article/rs-7473138\",\"identity\":\"rs-7473138\",\"version\":[\"v1\"]},\"buildId\":\"8U1c8b4HqxoKbykW_rLl7\",\"isFallback\":false,\"isExperimentalCompile\":false,\"dynamicIds\":[84888],\"gssp\":true,\"scriptLoader\":[]}","source_license":"CC-BY-4.0","license_restricted":false}