Mushroom Classification: Edible vs Poisonous

Overview

This project uses machine learning to classify mushrooms as edible or poisonous based on 22 categorical features describing physical characteristics. The dataset originates from the Audubon Society Field Guide to North American Mushrooms and was donated to the UCI Machine Learning Repository in 1987. Given the real-life consequences of misidentification, this problem is both challenging and impactful.

Problem Statement

Objective:
Build a classification model to predict whether a mushroom is edible (e) or poisonous (p) based on observable traits like odor, gill size, cap shape, and spore print color.

Dataset

• Samples: 8,124
• Features: 22 categorical features + 1 target label (class)
• Target Values:
  • e = Edible
  • p = Poisonous
    Note: The original "unknown edibility" category was combined with poisonous to ensure safety.
• Source: [https://www.kaggle.com/datasets/uciml/mushroom-classification)

Steps of Work Done

1. Data Loading & Exploration

• Loaded and inspected the dataset using pandas
• Counted total rows and columns
• Identified that all features are categorical
• Noted missing values: stalk-root had ~30.5% missing, marked as "?"

2. Target & Class Distribution

• Verified target class (class) is nearly balanced
  • Edible: 51.8%
  • Poisonous: 48.2%
• Plotted distribution for transparency

3. Feature Analysis

• Used pd.crosstab to find features highly correlated with the poisonous class
• Top predictors included:
  • odor
  • gill-size
  • spore-print-color
• Created bar plots and histograms to visualize feature separability

4. Data Cleaning

• Dropped veil-type (constant feature)
• Replaced missing stalk-root values with its mode
• One-hot encoded all categorical features using sklearn.OneHotEncoder with drop='first'
• Final shape: 8124 samples × 92 encoded features

5. Data Splitting

• Split into:
  • 80% training
  • 10% validation
  • 10% testing
• Used stratified sampling to maintain class balance

6. Model Training

• Chose Random Forest Classifier:
  • 100 trees
  • Class weight = "balanced"
  • Random seed for reproducibility

7. Evaluation Metrics

• Evaluated on both validation and test sets
• Reported:
  • Accuracy
  • Precision
  • Recall
  • F1-score
  • ROC AUC
• Confusion matrices visualized using seaborn

8. Feature Importance

• Extracted top 20 features by importance from the trained model
• Displayed using a horizontal bar chart

Project Structure

├── FinalProj.ipynb           
├── mushrooms.csv             
├── README.md