Valorant Match Analysis Tool

📝 Overview

This tool captures screenshots of Valorant matches, analyzes them in real-time, and predicts match outcomes based on extracted features. It includes live visualizations of win rates and detailed feature tables for analysis.

However, as the live processing is only consistent 90% of the time, this tool also offers an option to predict based on a screenshot instead of capturing screenshots and predicting live.

💪 Training

This model is only trained on the following features extracted from the image

Number of players alive
Health of each alive players
Number of ability points
Number of ults
Time since spike is planted (0 if spike is not planted)

Through the training process an accuracy of 95% can be achieved

🔍 Features

Real-time Analysis: Screenshots of the selected Valorant game window are captured and processed at regular intervals.
Feature Extraction: Extracts various game features such as number of players alive, ability counts, health stats, and presence of game elements like spikes.
Machine Learning Prediction: Uses a trained machine learning model (Random Forest) to predict match outcomes based on extracted features.
Live Visualization: Displays live updates of win rates over time and detailed feature tables using Matplotlib.

🛠 Installation

Clone the Repository:
Install the dependencies pip install -r requirements.txt

🚀 Usage

Using Live Version
- Run the script python3 use_live_forest.py
- Select Valorant Window:
  - A dialog will prompt you to select the Valorant game window from a list of currently open windows.
- Monitor Live Updates:
  - The application will start capturing and analyzing screenshots in real-time.
  - Win rates will be plotted dynamically, and feature tables will be updated continuously.
- Interpret Results:
  - View the live plot to monitor win rate trends.
  - Check the feature tables to understand the extracted data from each screenshot.
Using Single Image Version
- Change the image path at the bottom of the script (use_random_forest.py)
- Run the script python3 use_random_forest.py

🐛 Known Issues

Occasional delays or inaccuracies may occur in feature extraction depending on game window visibility and performance.
Slight inconsistencies in extraction of features when taking screenshots and predicting lvie

🔮 Future Enhancements

Gather more data to train the model (currently trained with 7000 images)
Add in additional features (e.g. agents played, guns used, economy)
Improve feature extraction to increase reliability
Enhance GUI for better user interaction and feedback.

🤝 Contributing

Contributions are welcome! If interested, please fork the repository and submit a pull request with your improvements. I can provide the dataset for training.