Mushroom Classification

Binary classification pipeline — edible vs poisonous — with EDA, feature engineering, four ML algorithms, confusion matrix, ROC & PR curves, and feature importance.

Overview

This project trains and evaluates four classification models on the UCI Mushroom dataset (8,124 samples, 23 categorical features). All preprocessing and training run on the backend; the frontend displays metrics and diagnostic plots without a page reload.

Models

Model	Notes
Random Forest	100 estimators
Gradient Boosting	100 estimators
Logistic Regression	max_iter=1000
Decision Tree	default depth

Features

• EDA plots — class balance, unique values per feature, stacked feature distributions, bivariate breakdown, Cramér's V correlation heatmap
• Metrics — accuracy, precision, recall, F1, AUC
• Diagnostic plots — confusion matrix, ROC curve, Precision-Recall curve, feature importance ranking
• Model comparison — side-by-side accuracy and F1 across all four models
• Caching — preprocessed data and trained models are cached in memory so re-running a model is instant

Dataset

UCI Mushroom dataset: 8,124 mushroom samples with 22 categorical features (cap shape, gill color, habitat, etc.) and a binary target (edible / poisonous). No missing values after imputation.

Tech Stack

• Backend: Flask, scikit-learn, pandas, numpy, matplotlib, seaborn, scipy
• Frontend: Chart.js 3.9.1, Vanilla JavaScript

Project Structure

mushroom_classification/
├── __init__.py                          # Flask blueprint, training pipeline, API routes
├── templates/
│   └── mushroom_classification/
│       └── index.html                   # Model selector, metrics display, plot panels
└── static/
    └── script.js                        # Fetch calls, Chart.js model comparison bar chart

API

Method	Path	Description
GET	/mushroom-classification/	Main page
GET	/mushroom-classification/run?model=<key>&force=<bool>	Train a model, return metrics + plots
GET	/mushroom-classification/compare	Train all models, return accuracy/F1 JSON
GET	/mushroom-classification/plots	Generate 5 EDA plots as base64 PNG

Model keys: random_forest, gradient_boosting, logistic_regression, decision_tree

Preprocessing

1. Mode imputation for missing values
2. Label encoding for all categorical features
3. 80/20 train/test split (stratified)
4. StandardScaler applied to encoded features