Forest CoverType Classification 🌲🌲

This project explores the Forest CoverType dataset to predict the type of forest cover from cartographic variables such as elevation, aspect, slope, soil type, and more.

It’s a multi-class classification problem with 7 cover type classes. Using ensemble methods, this project investigates how handling class imbalance and model choice impacts predictive performance.

---

📂 Dataset

• Source: Forest CoverType Dataset – Kaggle
• Classes: 7 types of forest cover.
• Characteristics:
  • Dataset already one-hot encoded, with no categorical variables present.
  • No duplicates or missing values.
  • Imbalanced classes: types 1 and 2 are majority classes, while 4 and 5 are underrepresented.

---

⚙️ Project Workflow

1. Data Preprocessing

   • Verified dataset cleanliness (no nulls, no duplicates).
   • Confirmed one-hot encoding of categorical variables.
   • Addressed class imbalance using SMOTE oversampling.

2. Modeling

   • Random Forest Classifier
     • Evaluated with confusion matrix, classification report, and feature importance visualization.
     • Validated with stratified cross-validation.
   • XGBoost Classifier
     • Applied the same evaluation (confusion matrix, classification report, feature importances).
     • Validated with stratified cross-validation.

3. Evaluation Metrics

   • Confusion Matrix
   • Classification Report (Precision, Recall, F1-score)
   • Stratified Cross-Validation

---

📊 Results

• Random Forest delivered the strongest performance with an average CV F1 of 0.93, making it the best fit for this dataset.
• XGBoost performed reasonably well with an average CV F1 of 0.8855, though slightly weaker than Random Forest.
• Feature importance analysis highlighted which environmental and soil-related factors most influenced forest cover classification.

---

🔑 Key Takeaways

• Class imbalance needed addressing for fairer evaluation — SMOTE oversampling proved useful.
• Random Forest outperformed XGBoost on this dataset, showcasing its robustness for high-dimensional, imbalanced data.
• Feature importance visualization provided ecological insights into which variables play the largest role in forest cover classification.