Stock Market Trend Discovery with Machine Learning

Azati Labs developed an AI-powered prototype that identifies stock market trends. The prototype leverages machine learning and sentiment analysis to evaluate how news articles impact stock price movements.

Discuss an idea

All Technologies Used

Python
Python
NLTK
NLTK
Keras
Keras
Gensim
Gensim
World2Vec
World2Vec

Motivation

To create a machine learning-based solution capable of analyzing news articles and predicting stock market trends based on narrative and sentiment analysis.

Main Challenges

Challenge 1
Lack of Data

The team lacked a sufficient dataset for training the models and had to manually collect, clean, and process data using custom web scrapers.

Challenge 2
Data Mapping

Unstructured text data required manual mapping and labeling. Two data entry specialists were involved to highlight key phrases influencing stock market trends.

Challenge 3
Sentiment Analysis Complexity

Training models to understand sentiment across different industries required extensive resources, making it difficult to generalize for various use cases.

Key Features

  • Custom data preprocessing scripts
  • Sentiment and narrative analysis using LSTM neural networks
  • Trend prediction with a probabilistic output
  • Scalable architecture for data processing

Our Approach

Sentiment and Narrative Analysis
The team focused on processing historical stock price changes alongside news events using LSTM neural networks for sentiment and narrative analysis.
Overcoming Data Limitations
To overcome data limitations, engineers built web scrapers and labeled data manually.
MVP Development
The solution was developed as an MVP with interconnected scripts for text preparation, machine learning, and result generation.
Pricing Policy Adjustment
Reviewed and refined the app’s pricing policy to retain customers during the improvement phase while critical bugs were addressed.

Project Impact

Outcome: The MVP achieved an average prediction accuracy of 65%. While not suitable for real-time API integration due to processing time, it demonstrated the feasibility of using news articles to predict stock trends.

Improvements: Enhanced data collection and cleansing techniques, increased prediction accuracy with larger datasets, and improved algorithm stability and scalability.

Ready To Get Started

This site is protected by reCAPTCHA and the Google Privacy Policy and Terms of Service apply.