A machine learning-based project that predicts stock price movements by combining traditional financial indicators with sentiment analysis derived from social media (Twitter) data. Built as a part of the Design Project at IIIT Vadodara.

Stock prices are influenced not only by financial indicators but also by public sentiment and news. In today's world, social media is a powerful source of sentiment, where the crowd’s opinion can affect market trends.

This project aims to predict stock price behavior using:

• Historical stock data (open, high, low, close, volume)
• Sentiment features derived from Twitter
• Statistical and machine learning models

• Predict stock price trends using machine learning models.
• Incorporate Twitter sentiment as a predictive feature.
• Compare performance with and without sentiment data.
• Improve on past models with better accuracy and generalization.

• Prior research shows sentiment analysis improves stock price prediction accuracy.
• Earlier works reported accuracies of ~70–80% using classical models.
• Our goal: surpass those benchmarks using combined feature engineering and optimized learning models.

• Collected daily stock features: Date, Open, High, Low, Close, etc.
• Trained models like:
  • Linear Regression → Error ≈ 1.325
  • Random Forest → Error ≈ 0.868

• Collected tweets using Tweepy API (past 24 hours).
• Preprocessed tweets for noise reduction.
• Used WordNet and SentiWordNet for sentiment scoring.
• Extracted features:
  • Positivity Score
  • Negativity Score

• Used a variation of Word2Vec to embed tweets.
• Generated weighted averages of tweet vectors across time spans.

• Tried various models; best results with:
  • LSTM (Long Short-Term Memory) neural network
• Configuration:
  • Activation: tanh
  • Optimizer: RMSprop
  • Epochs: 500

• Average model accuracy: ~90%
• LSTM outperformed traditional regression-based models.
• Models trained with sentiment features performed significantly better than those without.

• Public sentiment on platforms like Twitter plays a key role in stock price behavior.
• Combining financial and sentiment data improves prediction accuracy.
• Our model was able to reasonably predict stock behavior, if not exact prices.

• Train on longer historical windows (more than one day).
• Expand to multiple companies across sectors.
• Deploy as a live application or alert system.
• Use real-time sentiment streams for intraday predictions.

• Vikash Choudhary (201851144)
• Himanshu Bhadu (201851048)
• Ritik Rawat (201851102)
• Noorul Hasan Ali (201851078)

Indian Institute of Information Technology Vadodara
Government Engineering College, Sector-28, Gandhinagar, Gujarat - 382028

Dr. S.K. Patra
IIIT Vadodara

1. Illustrated Guide to LSTM’s and GRU’s – Michael Phi (Towards Data Science)
2. Stock Market Prediction Using Twitter Sentiment Analysis – Stanford
3. Machine Learning for Stock Prediction – Towards Data Science
4. Tweepy Documentation – Python Twitter API
5. SentiWordNet

📂 For more details, refer to the full project report: DesignProject-Report.pdf