Integrating Multimodal Data for Enhanced Stock Market Trend Prediction: A Deep Neural Network and Regression Approach

Abstract

The demand for accurate stock market trend prediction models has surged among financial traders, prompting the exploration of machine learning techniques to enhance predictive performance and reduce computational complexity. Traditional models often rely solely on historical stock data, which may not fully capture the intricate dynamics of financial markets. This research addresses this limitation by developing a model that utilizes Open, High, Low, and Close (OHLC) prices, integrating both classification and regression machine learning models to predict stock market trends. However, not all models are favourable for trend prediction and   therefore, we compare using different models to find the best performing model. The regression models tested include: Long Short-Term Memory (LSTM): Achieved 99.31% accuracy, Linear Regression: Achieved 98.85% accuracy, Decision Tree: Achieved 98.22% accuracy, Stochastic Gradient Descent (SGD): Achieved 97.63% accuracy, Temporal Convolutional Networks (TCN): Achieved 96.95% accuracy, K-Nearest Neighbors (KNN): Achieved 80.35% accuracy, Random Forest: Achieved 57.12% accuracy. The classification models evaluated include: Artificial Neural Networks (ANN): Achieved 97.28% accuracy, Stochastic Gradient Descent (SGD): Achieved 93.65% accuracy,  K-Nearest Neighbors (KNN): Achieved 89.53% accuracy, XGBoost: Achieved 87.01% accuracy,  Decision Tree: Achieved 86.1% accuracy, Random Forest: Achieved 85.06% accuracy, Support Vector Machine (SVM): Achieved 61.94% accuracy, AdaBoost: Achieved 58.38% accuracy, Naïve Bayes: Achieved 51.4% accuracy. Here, it is observed that the regression models are performing better in trend prediction when compared with classification models. However, it has also been noticed that ANN in classification; LSTM, and Linear regression in regression are performing better than other models in their categories