A COMPARATIVE ANALYSIS FOR STOCK PRICE PREDICTION USING IMPROVED EXPANSIVE DEEP LSTM MODEL Pankaj Rambhau Patil Amity School of Engineering & Technology, Amity University, Maharashtra, India patil.pankaj01@gmail.com Deepa Parasar Amity School of Engineering & Technology, Amity University, Maharashtra, India dparasar@mum.amity.edu Shrikant Charhate Amity School of Engineering & Technology, Amity University, Maharashtra, India scharhate@mum.amity.edu Abstract Due to a lack of clarity and flexibility, prediction leveraging ML models is not well fitted in many sections of commercial decision processes. Proposed model aim to employ deep learning strategy in the stock market pricing area to generate positive risk-adjusted price by analyzing previous transaction data and maintaining greater accuracy with a lower error rate. In this study, the deep learning approach is used, which is capable of handling time-series data. The results are obtained with evaluation of error rate metric MSE & RMSE which express how distant the data points are from the regression line. RMSE measures the dispersion of these residuals. It shows how concentrated the data is on the best fit line. This study compares a unique deep learning methodology with deep LSTM, GA and Harris Hawk optimization. As a part of this analysis results are observed and plotted for the various company stocks dataset, which clearly shows the effectiveness of proposed approach with reduced error rate. Keywords: Stock market prediction; recurrent neural network; deep LSTM; DEEP RNN; Deep Learning. 1. INTRODUCTION Commodities and wealth may provide us with a pleasant and secure lifestyle. It's no wonder that the research and forecasting of potential financial market values and forecasts has received so much interest. A number of forecasting approaches have been suggested and put into practice. Each strategy has benefits and drawbacks of its own. Also with development of internet trading in recent years, the stock market became choice for novice investors to generate substantial gains. Therefore it’s highly appealing if system model could correctly foresee market behavior so that investors may take appropriate decision about investment. Developing such a forecasting model is a challenging issue owing to the considerable implied volatility rules driving price movement. Neural networks (NNs) have become a highly essential tool for stock price forecasts due to their capacity to cope with ambiguous, imprecise, or inadequate data that fluctuates frequently in very short times [Schoenberg (1990)]. The purpose of this study is to highlight the problems that have to be explored in future research as well as to outline the main advantages and limitations of previous methodologies in RNN applications for stock markets. Certain benefits and limits were discovered after a comparative review of earlier research's methodology in respect to issue domains, data models, and results criteria. The stock exchange may be affected by a wide range of intricate events, such as economic cycles, warning strategies, interest rates, political ideologies, etc. There are many forecasting patterns, however most of them have advantages and downsides of their own. The underlying e-ISSN : 0976-5166 p-ISSN : 2231-3850 Pankaj Rambhau Patil et al. / Indian Journal of Computer Science and Engineering (IJCSE) DOI : 10.21817/indjcse/2022/v13i6/221306019 Vol. 13 No. 6 Nov-Dec 2022 1764