Stock Price Prediction using LSTM and ARIMA

Abstract

The stock market has always been a center of attention for investors. Tools that help in stock trend forecasting are in high demand as they help in the direct accession of profits. The more precise the results, is the higher chances of acquiring more profit. Factors such as politics, economics, and society impact the trends of the stock market. The analysis of stock trends can be performed using fundamental or technical analysis. The fundamental analysis comprises data from financial records, company assets, market shares economic reports, etc. Here analysis of the company's financial factors including the strategic initiatives, microeconomic indicators, and consumer behavior is conducted. Technical analysis is the interpretation of past and present prices with the help of which the probable future prices will be predicted. To plot stock market forecast various deep learning and machine learning algorithms are used. Out of which LSTM and ARIMA have proven to provide fairly accurate results. The papers presented before focused on the individual models and their components to provide forecasts. Their aim was the provide the predictions with the best-suited parameter values. The purpose of the paper is to offer the investors models which can work well with the data with appropriate parameter values. The aim is to provide both the models LSTM and ARIMA because of their capability to provide appropriate results with the help of technical analysis of the data set. The objective is to compare both the models and use the best one suited when it comes to a particular company data set. The data set used are the historic stock prices of the companies which include the open, close, high, low values. On which pre-processing is done which involves sorting the data, feature scaling, autocorrelation check, splitting it into training, and testing data sets. The results obtained show that the individual models work well when the data provided suits the model and appropriate parameter values are set. The accuracy of the models for each attribute is over 90%. Both the models can prove to be an asset to stock traders. The LSTM model provides better results when the data set is large and has fewer Nan values. Whereas, despite providing better accuracy than LSTM, the ARIMA model requires more time in terms of processing and works well when all the attributes of the data set provide legitimate values. Thus, the paper presented provides models which can visualize the future trend of the stock as graphs and give an overview of stock trend.