Enhanced Stock Market Prediction Using Hybrid LSTM Ensemble

Abstract

Stock market value prediction is the activity of predicting future market values so as to increase gain and profit. It aids in forming important financial decisions which help make smart and informed investments. The challenges in stock market predictions come due to the high volatility of the market due to current and past performances. The slightest variation in current news, trend or performance will impact the market drastically. Existing models fall short in computation cost and time, thereby making them less reliable for large datasets on a real-time basis. Studies have shown that a hybrid model performs better than a stand-alone model. Ensemble models tend to give improved results in terms of accuracy and computational efficiency. This study is focused on creating a better yielding model in terms of stock market value prediction using technical analysis, and it is done by creating an ensemble of long short-term memory (LSTM) model. It analyzes the results of individual LSTM models in predicting stock prices and creates an ensemble model in an effort to improve the overall performance of the prediction. The proposed model is evaluated on real-world data of 4 companies from Yahoo Finance. The study has shown that the ensemble has performed better than the stacked LSTM model by the following percentages: 21.86% for the Tesla dataset, 22.87% for the Amazon dataset, 4.09% for Nifty Bank and 20.94% for the Tata dataset. The model’s implementation has been justified by the above results.