A Future Trading System Using Ensemble Deep Learning

Abstract

Stock price prediction is a challenging research problem due to various dependencies like news headlines, Twitter, microblogs, and price history. News headlines and price history are the most significant features for predicting stock prices, which help investors decide to sell, buy, or hold that maximizes the profits. The manual analysis of large amounts of news and price history information is impossible. Hence, automation is required to summarize the available data for wise trading decisions. We propose an ingenious Future Trading System Using Ensemble Deep Learning (FTSEDL) to improve the accuracy of the prediction. The objective is to develop FTSEDL that improves the Deep learning predictive model with a new context-based clustering, CbCA, and information fusion for an accurate stock price forecast. The CbCA filters the required features from the news datasets to generate quality datasets through context-based clustering. Information fusion combines the two vital features of stock price prediction: news headlines and price history. The present research experimented with six stocks, BHARATIARTL, SBIN, PNB, NXTDIGITAL, TCS, and WIPRO, covering possible up, down, and sideways trends to improve the predictive model evaluation. The experiment data considered the stock news and price history data from 8thAugust 2016 to 31th March 2023, i.e., 2427 trading days. CbCA achieved a homogeneity of 0.95, Completeness of 0.98, and Silhouette coefficient of 0.88. The proposed model, FTSEDL, achieved an accuracy of 91.693.09, RMSE of 13.14, and MAPE of 0.02, which outperformed the models in the recent literature. The FTSEDL has many advantages when compared with contemporary literature. In addition, there is vast scope to apply this concept in real time.