Ensemble Machine Learning For Nifty-50 Price Forecasting And Trend Classification: A Flask-deployed Decision Support System

Abstract

Stock Market Forecasting Poses A Significant Challenge Due To The Non-linear, High-volatility Nature Of Financial Time Series. This Paper Presents An End-to-end Machine Learning Pipeline For Predicting NIFTY-50 Closing Prices And Next-day Directional Trends. The System Trains Random Forest (RF) And Decision Tree (DT) Regressors On Historical OHLCV Data Augmented With Engineered Technical Features (MA10, MA50, Daily Returns). A Fusion Mechanism Averages RF And DT Outputs To Produce A Stabilized Price Estimate. A Separate RF Classifier Outputs Categorical Trend Labels (UP/DOWN/NEUTRAL), Avoiding The Pitfall Of Inferring Direction From Regression Residuals. Experimental Results Show That The RF+DT Fusion Achieves An R² Of 0.9451, Outperforming Standalone RF (0.9312) And DT (0.8841) Regressors. The Trend Classifier Achieves 82.4% Accuracy And An F1-score Of 0.81. The Complete Pipeline Is Deployed As A Flask Web Application Supporting User Authentication, Interactive Prediction, Candlestick Visualization, CSV Upload, Live Data Fetch Via Yahoo Finance, PDF Report Export, And An Administrative Panel. The System Provides A Practical, Interpretable, And Deployable Solution For Short-term NIFTY-50 Decision Support.

Keywords