🌟 Forest Cover Type Prediction with Neural Networks 🌲

Welcome to the Forest Cover Type Classifier repository! This project
uses a TensorFlow-based neural network to predict forest cover types
from the UCI Covertype dataset. 🌳🔍 Whether you're into machine
learning, environmental data, or just love forests, this notebook has
you covered! We preprocess data, train a multi-layer perceptron,
evaluate performance, test random samples, and plot ROC curves for
insightful analysis. 🚀

📋 Project Overview

• Dataset: UCI Forest Covertype (581,012 samples, 54 features, 7
  classes like Spruce/Fir, Lodgepole Pine, etc.) 🌿
• Model: Sequential DNN with Dense layers, Dropout for
  regularization, and Softmax output. Trained with Adam optimizer and
  categorical cross-entropy. 🧠
• Key Features:
  • Data splitting & standardization 📊
  • Batch training with TensorFlow Datasets ⚡
  • Accuracy evaluation (~85% on test set) ✅
  • Random sample prediction testing 🎲
  • Multi-class ROC curve visualization 📈
• Tech Stack: TensorFlow, Scikit-learn, Matplotlib, NumPy 🛠️

🛠️ Installation

1. Clone the repo:

   git clone https://github.com/shervinnd/forest-cover-type-classifier.git
   cd forest-cover-type-classifier

2. Install dependencies (use a virtual environment like venv or conda):

   pip install tensorflow numpy matplotlib scikit-learn

3. Open the Jupyter Notebook:

   jupyter notebook covtype.ipynb

   Note: This was tested on Python 3.12 with GPU acceleration (T4).
   Ensure TensorFlow is GPU-enabled if needed! ⚙️

🚀 Usage

1. Run the Notebook: Execute cells sequentially to:
   • Import libraries 📚
   • Load & preprocess data (fetch_covtype, scaling, one-hot
     encoding) 🔄
   • Build & compile the model 🏗️
   • Train for 20 epochs (batch size 128) ⏱️
   • Evaluate on test set 📉
   • Test a random sample 🎯
   • Generate ROC curves for each class 📊
2. Customize:
   • Tweak hyperparameters like epochs, batch size, or layers in the
     parameters cell. 🔧
   • Run test_random_sample() multiple times for fun predictions!
     😄
3. Output Example:
   • Training logs show accuracy improving to ~81% on train, ~85%
     on validation.
   • ROC AUCs: High for most classes (e.g., 0.99+ for some)! 🌟

📊 Results & Insights

• Test Accuracy: ~85.23% 🎉
• Sample Prediction: Picks a random test instance, predicts cover
  type (e.g., Lodgepole Pine), shows probabilities, and checks
  correctness. ✅/❌
• ROC Curves: Visualizes model confidence per class -- great for
  multi-class imbalance analysis! 📈 (Plotted with Matplotlib)
• Pro Tip: Classes like Cottonwood/Willow might have lower AUC due to
  fewer samples. Experiment with oversampling! ⚖️

🤝 Contributing

We'd love your input! 🌍

• Fork the repo & create a pull request.
• Suggestions: Add more models (e.g., CNNs, XGBoost), hyperparameter
  tuning with Keras Tuner, or deployment with Streamlit. 💡
• Report issues or bugs via GitHub Issues. 🐛

📄 License

This project is licensed under the MIT License -- feel free to use,
modify, and share! 📜

Powered by Miracle ⚡ -- Exploring forests one prediction at a time!
🌲