Forecasting Stock Prices Using Time-Series Analysis, Regression Learning, and Deep Learning Models

Abstract

For an extended period, scholars have been crafting a dependable and precise forecasting model for predicting stock prices. Predictive models may meticulously and accurately anticipate future stock prices if they are properly created and improved, according to the literature. This paper presents a number of learning-based, econometric, and time series models for predicting stock prices. Here, the data from WIPRO, SBI, and APOLLO PHARMA from January 2000 to December 2021 was utilized to train and test the models in order to determine that algorithms worked in what industry. This study includes two Regression models (Random Forest and MARS), two deep learning-based models (basic RNN and LSTM), one econometric model (ARIMA), and one-time series model (Holt- Winters Exponential Smoothing). It has been demonstrated that LSTM is the best deep learning model and that MARS is the greatest machine learning model. But overall, MARS has shown to be the greatest-performing model in sales forecasting for all sectors: Information technology (using values from WIPRO), Finance (using values from SBI), and Health (using data from APOLLO PHARMA). Keywords— Time Series Analysis, Forecast stock price, Recurrent Neural Networks, LSTM, Random Forest, MARS.