Comparative Analysis of Machine Learning Algorithms for Predicting LQ45 Stock Index Prices

An essential metric for assessing the success of the country's capital markets is the LQ45 index, which is made up of 45 stocks with the biggest market capitalization and liquidity on thb e Indonesia Stock Exchange. Stock price prediction, particularly in volatile markets, remains complex challenge that benefits from advanced analytical approaches. While machine learning (ML) techniques have demonstrated significant promise in financial forecasting, comprehensive comparative evaluations across multiple algorithms and preprocessing strategies remain limited. In order to evaluate the predictive performance of nine machine learning algorithms Random Forest, Decision Tree, AdaBoost, Support Vector Classifier (SVC), XGBoost, Naive Bayes, K-Nearest Neighbors (KNN), Logistic Regression, and Artificial Neural Networks (ANN) in predicting the direction of movements of the LQ45 index, this study presents a structured comparative framework. The models are trained using a 10-year historical dataset, incorporating both continuous and binary representations of technical indicators. Three data preprocessing approaches are explored: raw trading data, unsmoothed indicators, and smoothed indicators. Accuracy, precision, recall, F1-score, and ROC AUC are all important factors in model evaluation. The findings show that when applied to continuous data with smoothed technical indications, Random Forest and XGBoost produce the best prediction results. For binary classification tasks, Naive Bayes emerges as the most effective model. These results demonstrate how important data representation and preprocessing in particular, smoothing are to enhancing the accuracy and robustness of models. Research aids in the creation of trustworthy, data-driven stock prediction tools that are suited for developing markets. Financial analysts, portfolio managers, and algorithmic traders looking to improve investment strategies through well-informed model selection and preprocessing design can benefit from the findings.