🏆 AI Model Arena Rank Predictor
An interactive Machine Learning Web Application built with Streamlit that predicts the leaderboard rank of AI models based on evaluation metrics like Rating, Variance, Vote Count, and Year.
🖼️ Application Preview & UI
Enter numerical model metrics in the interactive Streamlit dashboard to instantly estimate leaderboard rank.
🔗 Live Application
You can access and test the deployed application directly here: 👉 Click Here to Launch Live Demo
📌 Project Architecture & Model Evolution
1. Initial Approach: Deep Learning (Sequential Neural Network)
Initially, a multi-layer Deep Neural Network (Sequential) was trained using Keras, incorporating Dense, BatchNormalization, and Dropout layers. However, due to a relatively small dataset size, the deep learning approach suffered from lower performance and overfitting compared to traditional Machine Learning algorithms.
2. Optimized Approach: Tree-Based Machine Learning
To achieve higher accuracy and better generalization, the pipeline was transitioned to Machine Learning algorithms. Ensemble methods—specifically Extra Trees Regressor ($R^2 \approx 0.9858$) and Random Forest Regressor ($R^2 \approx 0.9794$)—outperformed all other models by a significant margin.
📊 Model Evaluation & Benchmarks
Here is the comprehensive performance metric comparison across all evaluated Machine Learning models:
| Model Name | $R^2$ Score | MSE | MAE | RMSE |
|---|---|---|---|---|
| Extra Trees Regressor 🏆 | 0.985858 | 114.65 | 5.04 | 10.71 |
| Random Forest Regressor | 0.979452 | 166.58 | 6.00 | 12.91 |
| Decision Tree Regressor | 0.962737 | 302.08 | 6.48 | 17.38 |
| K-Neighbors Regressor | 0.949303 | 410.98 | 9.82 | 20.27 |
| Gradient Boosting | 0.708248 | 2365.16 | 34.72 | 48.63 |
| Linear Regression | 0.527966 | 3826.66 | 49.34 | 61.86 |
| Ridge Regression | 0.527965 | 3826.67 | 49.34 | 61.86 |
| Lasso Regression | 0.527865 | 3827.48 | 49.37 | 61.87 |
| ElasticNet | 0.516176 | 3922.24 | 50.52 | 62.63 |
Key Insight: Extra Trees achieved the best performance with an $R^2$ score of ~98.58% and a Low Mean Absolute Error (MAE) of 5.04.
🛠️ Tech Stack & Dependencies
- Frontend / UI: Streamlit
- Machine Learning: Scikit-Learn
- Deep Learning (Experimental): Keras / TensorFlow
- Data Manipulation: Pandas, NumPy
- Model Serialization: Joblib
📁 Repository Structure
├── app.py # Main Streamlit Web Application
├── requirements.txt # Python Dependencies
├── ai_model_arena_rankings/ # Project Directory
├── .gitignore # Git Ignore File
├── .gitattributes # Git Attributes File
├── UI.png # Streamlit UI Screenshot
├── ML_model.pkl # Best Trained ML Model (Extra Trees / Random Forest)
├── scaler.pkl # Fitted StandardScaler Object
├── columns.pkl # Feature Column Definitions
├── notebook.ipynb # Model Training & EDA Notebook
├── requirements.txt # Python Dependencies
└── README.md # Project Documentation
git clone https://github.com/amirsohail100/AI-Model-Arena-Rank-Predictor.git
cd AI-Model-Arena-Rank-Predictor
streamlit run app.py
pip install -r requirements.txt
📄 License
This project is licensed under the MIT License.
📝 Author
👤 Amir Sohail
AI Model Arena Rank Predictor is a Streamlit web application designed to forecast model ranks based on ratings, variance, vote counts, year, and text descriptions. Built with Python, Scikit-Learn, Keras, and Joblib, it provides seamless real-time predictions using pickled scalers, tokenizers, and ML/DL models.