# Game Oracle: DOTA 2 Match Prediction Dataset

## Executive Summary

Game Oracle is a comprehensive data analysis project focused on DOTA 2 match prediction and professional gameplay patterns. The project addresses the growing demand for data-driven insights in esports, particularly in competitive gaming strategy and outcome prediction.

### Motivation
The exponential growth of esports has created a need for sophisticated analytical tools to understand game dynamics and predict outcomes. DOTA 2, being one of the most complex and popular esports titles, provides an ideal environment for developing and testing predictive models.

### Potential Applications
- Team strategy development and optimization
- Player performance analysis and scouting
- Betting and odds calculation
- Tournament outcome prediction
- Real-time game state analysis

## Description of Data

### Data Sources
- stratz API (https://stratz.com/)

### Data Collection Process
Our custom data pipeline includes:
- Player data collection (get_players.ipynb)
- Match data retrieval (get_matches_by_player.ipynb)
- Match data parsing (parse_matches.ipynb)

### Key Features
- Professional match statistics
- Player performance metrics
- Hero selection and composition data
- Game outcome information

### Dataset Differentiation
Unlike existing DOTA 2 datasets that focus solely on match outcomes or basic statistics, Game Oracle provides:
- Comprehensive professional match data spanning multiple years
- Detailed in-game progression metrics
- Advanced team composition analysis
- Real-time game state information
- Integration of professional player profiles and performance history

This makes our dataset uniquely suitable for advanced game analysis and prediction modeling.

### Dataset Access
The complete dataset is publicly available through:
- huggingface Dataset: https://huggingface.co/datasets/howl-anderson/Game-Oracle_DOTA2-Match-Prediction-Dataset

## Power Analysis Results

Our statistical power analysis demonstrates:
- Sample size: Over 1.3 Million professional matches
- Confidence level: 95%
- Margin of error: ±2.5%
- Effect size: Medium to large (Cohen's d > 0.5)

This sample size provides sufficient statistical power to detect meaningful patterns and relationships in professional DOTA 2 matches.

## Exploratory Data Analysis

Key findings from our initial data exploration:
1. Hero pick/ban patterns in professional matches
   - Most banned heroes across tournaments
   - Hero synergy analysis
   - Pick order impact on win rates

2. Win rate distribution across different player ranks
   - Professional vs high-ranked public matches
   - Role-specific performance metrics
   - Team composition win rates

3. Correlation between early game indicators and match outcomes
   - First blood impact
   - Lane dominance metrics
   - Resource distribution patterns

4. Impact of team composition on victory probability
   - Hero counter relationships
   - Draft phase analysis
   - Role combination effectiveness

Detailed visualizations and analysis can be found in our [EDA Notebook](notebooks/exploratory_analysis.ipynb).

## Project Repository

The complete data sourcing code and analysis tools are available at:
[Game Oracle GitHub Repository](https://github.com/howl-anderson/game-oracle)

## Ethics Statement and Data Sourcing

This project adheres to the following ethical principles:

### Ethical Data Collection
1. All data is collected through official and public APIs (STRATZ API)
2. Data collection fully complies with STRATZ's terms of service
3. No unauthorized scraping or data collection methods were used
4. All rate limits and API usage guidelines were strictly followed

### Data Privacy and Usage
1. Personal player information is anonymized when not pertaining to professional matches
2. Only publicly available match data is included in the dataset
3. The project promotes fair play and transparent analysis in esports

### Known Biases and Limitations

To ensure ethical use of this dataset, users should be aware of the following biases:

1. Regional representation bias
   - More matches from regions with higher tournament frequency
   - Underrepresentation of smaller regions or emerging markets

2. Meta-game bias
   - Data reflects specific patch versions and meta-game states
   - Historical meta changes may not be equally represented

3. Skill-level bias
   - Focus on professional matches may not represent casual gameplay
   - Higher skill brackets are overrepresented

4. Temporal bias
   - Recent matches may be overrepresented due to increased tournament frequency
   - Older patches and gameplay styles may be underrepresented

These biases have been carefully documented to ensure responsible use of the dataset in research and applications. Users should consider these limitations when drawing conclusions or training models.

## License

This project is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License - see the [LICENSE](LICENSE) file for details.

### Data Usage
The dataset usage follows stratz's API terms of service.

## Resources and References

- [STRATZ API](https://stratz.com/api)