PERFORMANCE EVALUATION AND OPTIMIZATION OF A MACHINE LEARNING-BASED PREDICTIVE FRAMEWORK FOR STOCK MARKET TREND ANALYSIS USING REAL-WORLD FINANCIAL DATA

Abstract

The stock market is a nonlinear, dynamic and unpredictable market, making accurate forecasting a difficult task. In this paper, a machine learning approach for stock market trend prediction is proposed based on real-world financial data and technical indicators. The framework includes a comprehensive data preprocessing pipeline, feature engineering, and hyperparameter optimization, all aimed at enhancing the accuracy of predictions and the robustness of the model. The historical stock market information is loaded, and then converted to useful features such as moving averages, Relative Strength Index (RSI), Moving Average Convergence Divergence (MACD), Bollinger Bands, trading volume, and daily returns. Various machine learning models, including Logistic Regression (LR), Decision Tree (DT), Random Forest (RF), Support Vector Machine (SVM), XGBoost and Long Short-Term Memory (LSTM) are tested and compared with classical performance metrics, such as precision, recall, F1-score, cross-validation accuracy and Area under the Receiver Operating Characteristic (ROC) curve (AUC). The experimental results show that the proposed optimized predictive framework outperforms the conventional machine learning approaches, with an accuracy of 97.13%, an F1-score of 0.970 and an AUC of 0.991. The framework also showed very good performance with 5-fold cross validation, which means good generalizing ability for new market situations. The methodology outlined here is efficient and reliable to offer a decision support tool for investors, financial analysts, and automatic trading systems to improve stock market predictions and financial decisions in changing markets.