Dota 2 Match Predictor

Project Details

This was my thesis project completed in May 2025, combining web development with machine learning to predict Dota 2 match outcomes.

System Architecture:

• Frontend: Built with HTML, CSS, JavaScript, PHP and SQL for data visualization
• Backend API: Python with Flask framework to serve predictions
• ML Pipeline: Utilized Hugging Face transformers for feature processing
• Prediction Model: Stacked BiLSTM architecture trained on professional match data

Key Features:

• Real-time match prediction during drafting phase
• Win probability visualization throughout the match
• Historical performance analysis of players/teams
• Hero matchup advantage/disadvantage indicators

Technical Highlights:

• Processed over 102,000+ professional matches for training
• Achieved 97%~ prediction accuracy on test set
• Implemented custom feature engineering for Dota-specific metrics
• Optimized model for low-latency predictions during live matches