A Multivariate Hybrid Deep Learning Model for Financial Stock Market Price Prediction

Abstract

The stock market is a volatile part of the global financial market. Millions of financial transactions occur every second in the former, representing billions of dollars in value; therefore, the assessment and prediction of stock prices constitute an important area of research. Economic experts, traders, and investors seek a method and model that can assist them in predicting stock price trends and identifying better strategies for making informed investments. Therefore, Stock Price Prediction (SPP) is crucial for scholars from both technical and financial fields. Efficient price and next-day price forecasting can significantly impact the investment decision concerning a portfolio of equity instruments. The development of Deep Learning (DL) has resulted in several innovations in stock market forecasting. This study presents a Multivariate Hybrid Deep Learning Driven Stock Market Price Trend Prediction (MHDL-SMPTP) model. The proposed next-day price forecasting model accurately predicts short-term stock market price trends to enhance price forecasting using advanced techniques. For analysis, z-score normalization methodology is utilized for cleaning and processing unstructured data into an organized form, followed by feature selection using Ant Colony Optimization (ACO). Furthermore, a hybrid Convolutional Neural Network (CNN) combined with a Gated Recurrent Unit (CNN-GRU) network is employed for prediction. Finally, parameter fine-tuning is performed using the Horned Lizard Optimizer Algorithm (HLOA) model. The proposed model is evaluated utilizing a historical stock market dataset. The short-term SPP results indicate that the proposed model achieved the lowest error values, with an RMSE of 0.4487, an MAE of 0.4322, and an MAPE of 0.5282.