ANALYZING THE INFLUENCE OF BIG DATA HETEROGENEITY ON STOCK MARKET PREDICTION THROUGH INFORMATION FUSION AND DEEP LEARNING MODELS

Abstract

In recent years, with the rise in the availability of heterogeneous big data sources such as stock price data, financial news, social media posts, Google Trends, and macroeconomic data, stock market forecasting (SMP) has entirely changed. However, a unified integration of these heterogeneous datasets for successful forecasting remains a long-standing problem. This work proposes a deep learning–based framework applying information fusion for fusing diverse financial and social data towards more accurate prediction. The method entailed systematic data preprocessing, sentiment analysis, feature extraction, normalization, and fusion and subsequent classification with Long Short-Term Memory (LSTM) and Convolutional Neural Network (CNN) models. Experimental validation indicates that the best prediction accuracy of the LSTM model was 99% on stock market data compared CNN and even previous reports. The results confirm that data fusion of heterogeneous data sets considerably enhances predictive accuracy to enable superior and wiser financial decision-making.