Implementation of Stacking Technique Combining Machine Learning and Deep Learning Algorithms Using SMOTE to Improve Stock Market Prediction Accuracy

Abstract

This study introduces a stacking technique that integrates machine learning (ML) and deep learning (DL) algorithms to enhance the accuracy of stock market trend predictions. The stacking model utilizes XGBoost and Random Forest as base models from the ML domain, while Logistic Regression and LSTM (Long Short-Term Memory) function as meta models to optimize predictive accuracy. A significant challenge in stock market data is class imbalance, where certain trends, such as stock price drops, are underrepresented. To mitigate this, we applied the Synthetic Minority Over-sampling Technique (SMOTE) to generate synthetic data for the minority class. This approach helps the model better capture patterns from the underrepresented data while preserving essential information from the majority class. The implementation of SMOTE, coupled with the stacking technique, yielded a substantial improvement in prediction accuracy. The results showed that the Random Forest algorithm achieved an accuracy of 85% with precision, recall, and F1-score all at 85%, while XGBoost and Logistic Regression achieved accuracies of 82% and 81% respectively. For the deep learning models, LSTM reached an accuracy of 83%, while the Stacking Meta Model with LSTM achieved an accuracy of 83% with slightly better precision and recall at 84%. The stacking model, with Logistic Regression as the meta model, ultimately achieved the highest accuracy of 86%, outperforming individual models such as SVM (Support Vector Machine), LSTM, Random Forest, and Logistic Regression (LR). These findings demonstrate the efficacy of combining SMOTE with stacking to address data imbalance and improve stock market predictions. The novelty of this study lies in the integration of advanced ML and DL models within a stacking framework to handle class imbalance in financial datasets. Future research will explore the deployment of this model in a real-time web-based application to support investor decision-making in stock market trend analysis.