Paper Title
STOCK MARKET TREND ANALYSIS USING DATA SCIENCE TECHNIQUES: A COMPARATIVE STUDY OF LINEAR REGRESSION AND LSTM MODELS

Abstract
Stock forecasting in the market is still an uphill task due to the non-linearity, stochasticity, and volatility inherent with the financial time-series data. In many market dynamics, the complex temporal dependence that is often intrinsic in such dynamics cannot be modeled using conventional statistical techniques. The study will provide a methodical analysis of a conventional Linear Regression model versus a current deep-learning network, which is the Long Short-Term Memory (LSTM) model, used to predict stock price trends. The historical price series were compiled and underwent extensive preprocessing, and engineered features of returns on a daily basis, rolling volatility indexes and moving averages were built as predictive power enhancers. Both model families had been got schooled and assessed following the same experimental protocol, utilizing a seven-day predictive cyclic forecasting analysis modified to brief predictions. Through empirical analysis, it was found that the LSTM model significantly works better when compared with Linear Regression on the most important metrics such as mean squared error (MSE), root-mean-square error (RMSE), and directional accuracy. In addition, a hybrid assessment scheme which is volatility-regime-based is proposed in a bid to test the model efficacy in different market regimes, and thus provide practical information to trading-strategy optimization and risk-management schemes. Keywords - Deep Learning, Directional Accuracy, Financial Analytics, Linear Regression, LSTM, Rolling Volatility, Stock Market Prediction, Time-Series Forecasting, Trading Strategy, Volatility Regimes.