📌 Valorant Gameplay Analysis & Model Comparison

📝 Project Overview

This project analyzes Valorant gameplay data and evaluates different machine learning models to predict player performance. The goal is to understand key factors that influence match outcomes.

🎮 Dataset

• Source: Valorant match statistics
• Columns:
  • Player ID, Agent, KDA, Win/Loss, Map, Rank, Headshot %, Damage per Round
  • Rounds Won, Rounds Lost, Economy, Spike Plants, Spike Defuses

⚙️ Methodology

1. Data Preprocessing

   • Handling missing values
   • Label encoding the output
   • Encoding categorical data (e.g., Agents, Maps)

2. Exploratory Data Analysis (EDA)

   • Distribution of player statistics
   • Impact of different agents on performance
   • Correlation between rank and gameplay metrics
   • Implemented for loops to streamline the plotting of Countplots, Histplots, and Barplots across multiple features instead of using separate blocks

3. Model Implementation

   • Compared XGBoost, Random Forest, and CatBoost using a for loop for efficiency
   • Performed hyperparameter tuning and k-fold cross-validation for better model performance

4. Model Evaluation

   • Accuracy and precision metrics
   • Classification Study
   • Feature Importance analysis (Not Updated in Code. Basic Fearure Engineering Was Done)
   • Previously, Random Forest had the best accuracy at 78 percent, but after optimization, CatBoost achieved 85 percent accuracy.

5. Dimensionality Reduction

   • Attempted dimensionality reduction using LDA, but it did not provide significant improvements. Hence, it was not included in the final project scope.

🏆 Conclusion

• CatBoost demonstrated the highest accuracy and is the most suitable model for this dataset
• Hyperparameter tuning played a crucial role in enhancing performance, highlighting the importance of fine-tuning machine learning models
• Dimensionality reduction methods like LDA may not always be beneficial, and their effectiveness depends on the dataset and feature distribution (Removed from code)
• Using for loops for visualization and model comparisons increased efficiency and improved code readability
• Future improvements could include testing additional models, feature engineering, and exploring alternative dimensionality reduction techniques

🚀 Feature Scaling & Deep Learning Experiment

I have performed feature scaling and deep learning separately in a Google Colab notebook. However, the results did not show a significant improvement nor outperformed CatBoost's accuracy of 85 percent.
I would love to hear your suggestions on improving the model performance.

🛠️ Installation & Usage

# Install dependencies
pip install pandas numpy matplotlib seaborn scikit-learn

# Clone the repository
git clone <repo_url>
cd <repo_folder>

# Open Jupyter Notebook
jupyter notebook valorant-analysis-models-comparison.ipynb