Predicting Stock Market Trends Using Machine Learning: A Sector-Wise Comparative Study of ARIMA, Linear Regression, and LSTM Models on NSE-Listed Companies

Abstract

-This study, titled “Predicting Stock Market Trends Using Machine Learning,” examines stock price behaviour across five major sectors of the Indian stock market — Information Technology, Banking, Energy, Automobile, and Pharmaceuticals — using historical daily price data of ten companies listed on the National Stock Exchange (NSE). The research problem addressed is the difficulty of achieving accurate stock trend prediction using conventional forecasting methods in a volatile market environment. The objectives were to evaluate stock price trends across the selected sectors, compare prediction performance across companies, develop forecasting models using ARIMA, Linear Regression, and Long Short-Term Memory (LSTM) networks, and identify the most effective technique. Secondary OHLCV data was collected from the NSE, cleaned, and chronologically split 80:20 for training and testing. Trend, volume, and volatility indicators were computed for all ten companies, while detailed model development was carried out for Tata Consultancy Services (TCS) as a representative case. Linear Regression used Open, High, Low, and Volume as predictors of the Closing price; ARIMA was applied to the differenced Closing price series; and the LSTM network was trained on scaled Closing price sequences using a sixty-day look-back window. Performance was assessed using MAE, RMSE, and R². The findings reveal clear sectoral divergence, with Information Technology and Banking stocks exhibiting bearish trends while Energy, Automobile, and Pharmaceutical stocks showed upward momentum. Linear Regression outperformed both ARIMA and LSTM, yielding an R² of 0.998 compared with 0.959 for LSTM and a negative value for ARIMA, resulting in a SELL recommendation for TCS. The study concludes that the most suitable forecasting technique depends on the nature of the input data and forecast horizon, and that model outputs should support, rather than replace, broader investment judgement.