Assignment 2 – Multi-Model Classification with Streamlit

---

1) Problem Statement

Build a Machine Learning model that predicts whether a mushroom is poisonous or edible based on its physical and environmental attributes. The goal is to help identify potentially harmful mushrooms early so safer decisions can be made while handling or consuming them.

This application evaluates multiple classification models to determine the outcome:

• e → Edible Mushroom
• p → Poisonous Mushroom

---

2) Dataset Description

This project uses a mushroom classification dataset containing real-world style observations of mushroom specimens. The dataset includes cap, gill, stem, veil, ring, and habitat-related features that are highly useful for predicting whether a mushroom is edible or poisonous.

Dataset Source

The dataset used in this project is available locally in this repository at: data/mushroom.csv

Dataset Overview

• Total Records: 12,214
• Total Columns: 21
• Input Features: 20
• Target Column: class

Attribute Details

• cap-diameter: diameter of the mushroom cap
• cap-shape: shape of the cap
• cap-surface: texture of the cap surface
• cap-color: color of the cap
• does-bruise-or-bleed: whether the mushroom bruises or bleeds
• gill-attachment: type of gill attachment
• gill-spacing: spacing between gills
• gill-color: color of the gills
• stem-height: height of the stem
• stem-width: width of the stem
• stem-root: root characteristic of the stem
• stem-surface: texture of the stem surface
• stem-color: color of the stem
• veil-type: type of veil present
• veil-color: color of the veil
• has-ring: whether a ring is present
• ring-type: type of ring
• spore-print-color: color of the spore print
• habitat: natural habitat of the mushroom
• season: season in which the mushroom appears
• class: mushroom class (e = edible, p = poisonous)

2.3 Train/Test split used

Data is split using stratified train-test split (test_size=0.2, random_state=42):

• Train set: 9,771 rows
• Test set: 2,443 rows

Data files:

• data/mushroom_train.csv
• data/mushroom_test.csv

---

3) Models Used and Evaluation Metrics

3.1 Models implemented

• Logistic Regression (model/logistic_regression.py)
• Decision Tree (model/decision_tree.py)
• KNN (model/knn.py)
• Naive Bayes (model/naive_bayes.py)
• Random Forest (model/random_forest.py)
• XGBoost (model/xgboost_model.py)

3.2 Evaluation metrics

All 6 models are evaluated on:

1. Accuracy
2. AUC Score
3. Precision
4. Recall
5. F1 Score
6. Matthews Correlation Coefficient (MCC)

---

4) Comparison Table (All 6 Models)

The following values are populated from metrics_comparison.csv:

ML Model Name	Accuracy	AUC	Precision	Recall	F1	MCC
Logistic Regression	0.8420	0.9184	0.8403	0.8438	0.8411	0.6841
Decision Tree	0.9947	0.9948	0.9945	0.9948	0.9946	0.9892
KNN	0.9992	1.0000	0.9992	0.9992	0.9992	0.9983
Naive Bayes	0.6013	0.8357	0.7550	0.6402	0.5666	0.3782
Random Forest (Ensemble)	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000
XGBoost (Ensemble)	0.9476	0.9913	0.9462	0.9482	0.9471	0.8944

---

5) Observations on Model Performance

ML Model Name	Observation about model performance
Logistic Regression	Good baseline performance and balanced precision/recall; lower than tree/ensemble methods on this dataset.
Decision Tree	Very strong performance across all metrics; captures non-linear splits effectively.
KNN	Near-perfect scores, indicating strong local separability of classes after preprocessing.
Naive Bayes	Lowest overall scores; conditional independence assumption appears less suitable for this data distribution.
Random Forest (Ensemble)	Best overall performance (perfect metrics in current run), indicating excellent robustness and generalization on this split.
XGBoost (Ensemble)	Excellent AUC and MCC; slightly below Random Forest/KNN but still high-performing and reliable.

---

6) Streamlit App Features

The Streamlit app (app.py) includes all required components:

• ✅ Dataset upload option (CSV)
• ✅ Model selection dropdown (multiple models)
• ✅ Display of evaluation metrics
• ✅ Confusion matrix / classification report

Additional implemented capabilities:

• Prediction preview table
• Downloadable prediction CSV
• Model comparison table and metric visualization

Run locally:

streamlit run app.py

---

7) Repository Structure (Required)

project-folder/
├── app.py
├── requirements.txt
├── README.md
├── README_ASSIGNMENT.md
├── train.py
├── metrics_comparison.csv
├── data/
│   ├── mushroom.csv
│   ├── mushroom_train.csv
│   ├── mushroom_test.csv
│   └── preprocess_data.py
└── model/
    ├── logistic_regression.py
    ├── decision_tree.py
    ├── knn.py
    ├── naive_bayes.py
    ├── random_forest.py
    ├── xgboost_model.py
    └── saved_models/
        ├── logistic_regression.pkl
        ├── decision_tree.pkl
        ├── knn.pkl
        ├── naive_bayes.pkl
        ├── random_forest.pkl
        └── xgboost.pkl

---

8) Requirements and Execution

Install dependencies:

pip install -r requirements.txt

Train all models and regenerate metrics:

python train.py

Run Streamlit app:

streamlit run app.py

---

9) Student Details

• Name: Sumanth_T_P
• BITS ID: 2025AA05544
• Streamlit App Link: https://ml-models-app-sumanth-tp-2025aa05544.streamlit.app/