A Hybrid Stock Price Forecasting System Using Arima, Lstm, And Sentiment Analysis

Abstract

The Accurate Forecasting Of Stock Prices Remains One Of The Most Persistent And Challenging Problems In Quantitative Finance, Owing To The Inherently Noisy, Nonstationary, And Nonlinear Nature Of Financial Time Series. Traditional Forecasting Methods Struggle With The Dual Challenges Of Capturing Both Linear Periodicities And Nonlinear Behavioral Dynamics Within A Unified Framework. This Paper Presents A Combined Approach To Enhance Stock Price Forecasting Accuracy. The Proposed Model Integrates Three Key Components: Autoregressive Integrated Moving Average (ARIMA) For Effectively Modeling Linear Patterns And Repeating Cycles In Stock Price Data; Long ShortTerm Memory (LSTM) Networks For Identifying Complex, Longterm Dependencies In Stock Price Movements; And Market News Sentiment Analysis To Capture Investor Behavior And Emotional Influences. By Synergistically Combining Statistical Timeseries Modeling, Deep Learning, And Sentimentdriven Insights, The Approach Aims To Overcome The Limitations Of Individual Methods. Experimental Evaluation On Three Largecap Stocks (AAPL, JPM, TSLA) Over A 12month Testing Period Demonstrates That The Hybrid Model Achieves A 41.9% Reduction In RMSE Compared To Standalone ARIMA And A 30.1% Reduction Compared To Standalone LSTM, With Directional Accuracy Improving From 52.9% And 59.9% To69.5%. The Hybrid Framework Provides More Robust And Precise Stock Price Predictions, Supporting Better Informed Financial Decision Making.

Keywords