📈 Fintel: News-Driven Stock Forecasting Prototype

Overview

Fintel is a proof-of-concept machine learning system designed to predict short-term stock price movements (Up, Down, or Neutral) by leveraging both historical price data and recent news sentiment.

Features

• Multi-Modal Data Integration: Combines numerical stock with latest financial news.
• Sentiment Analysis: Utilizes VADER (Valence Aware Dictionary and sEntiment Reasoner) to extract sentiment scores from news headlines and descriptions.
• Machine Learning Prediction: Employs an XGBoost Classifier to predict the next day's stock movement direction (Up, Down, Neutral).
• Web-Based GUI: A simple, intuitive graphical user interface built with Streamlit for easy interaction and demonstration.

Click for Video Demo

How it Works (Under the Hood)

1. Data Acquisition:
   • Stock Prices: Uses yfinance to fetch historical data for the stock.
   • News Articles: Connects to NewsAPI.org to retrieve recent news headlines and descriptions related to the stock.
2. Feature Engineering:
   • Numerical Features: Calculates lagged price data (e.g., previous day's close) and common technical indicators (e.g., Simple Moving Averages, RSI).
   • Textual Features: Processes news articles to extract sentiment scores using VADER. These are then aggregated daily (e.g., average daily sentiment, count of positive/negative news).
   • Data Fusion: Merges the daily stock features with the aggregated daily news features.
   • Target Definition: Transforms the next day's predictions as: "Up", "Down", or "Neutral".
3. Model Training:
   • An XGBoostClassifier is trained on the combined historical features to predict the Target_Direction.
   • The model and the StandardScaler (used for feature scaling) are saved for later use.
4. Prediction & Advice Generation:
   • When a prediction request is made, the system fetches the latest stock prices and news.
   • These real-time inputs are transformed into features using the same preprocessing steps and the saved StandardScaler.
   • The trained XGBoost model makes a prediction.
   • The SHAP advice model is currently being implemented.
5. Web GUI: Streamlit hosts an interactive interface that allows users to input a stock ticker (currently fixed to AAPL in the backend logic for this prototype) and view the prediction and advice.