Mushroom Classification using Multiple ML Models

A machine learning project focused on mushroom classification using multiple models and comparing their performance. The project includes comprehensive data preprocessing, feature engineering, and model evaluation using cross-validation techniques.

• Loading and splitting semi-colon separated dataset
• Handling 21 different features including:
  • Categorical: cap shape, surface, color, gill properties
  • Continuous: stem dimensions, cap diameter
• Target variable: mushroom classification

• Categorical Features Processing:

  • One-hot encoding for 17 categorical features
  • Feature label generation
  • Conversion to sparse matrix format

• Continuous Features Processing:

  • Standard scaling for numerical features
  • Normalization of stem dimensions and cap diameter
  • Feature concatenation for final dataset

• Implemented k-fold cross-validation (k=10)
• Consistent evaluation across all models
• Robust performance measurement

1. K-Nearest Neighbors (KNN)

   • Tested k values from 1 to 10
   • Best performance at k=15
   • Accuracy: 100%

2. Support Vector Machine (SVM)

   • Linear kernel implementation
   • C=1 hyperparameter
   • Accuracy: 88%

3. Decision Tree

   • Default parameters
   • Accuracy: 100%

4. Gaussian Naive Bayes

   • Probabilistic classifier
   • Accuracy: 60%

5. Neural Network

   • MLP Classifier
   • Alpha=1, max_iterations=1000
   • Accuracy: 100%

• Top Performing Models:
  • KNN (k=15): 100%
  • Neural Network: 100%
  • Decision Tree: 100%
• Mid-Range Performance:
  • SVM: 88%
• Lower Performance:
  • Gaussian NB: 60%

• Three models achieved perfect classification
• SVM showed good but not optimal performance
• Gaussian NB struggled with the feature space

• Pandas for data manipulation
• Scikit-learn for ML models
• NumPy for numerical operations
• Matplotlib for visualization

• Modular preprocessing pipeline
• Cross-validation implementation
• Model training and evaluation
• Performance visualization

• Feature importance analysis
• Hyperparameter tuning
• Ensemble methods implementation
• Additional data collection
• Model optimization for speed