Stock Prices Prediction Using Machine Learning

Abstract

Precise prediction of the stock market is incredibly difficult due to how volatile and non-linear it is. The growth of artificial intelligence and improvements in processing power have increased the accuracy of programmed methods of prediction in predicting stock values. In this paper, we used Multilayer Linear-Regression, Convolutional Neural Network (CNN), and long short-term memory (LSTM) algorithms to analyze the price trends over different time periods to predict the closing price of five companies, operating in different sectors. The final features we used were open, high, low, and close prices (OHLC), which were chosen using data pre-processing techniques. The dataset we used was made up of daily prices from 3 November 2012 to 3 November 2022. The number of previous days that would be required to predict the current day’s closing price is known as the sequence length. Then, using a different set of information, we adjusted this length and evaluated the accuracy. Our models performed the best for a sequence length of 5 and LSTM outperforms other models for each company’s dataset with different sequence lengths.