Decision Tree Classification — Interpretability Demo

Overview

This repository contains a recruiter-facing Decision Tree classification demo built with Python, scikit-learn, and Streamlit.

The project uses the Adult Income dataset to predict whether an input profile belongs to the <=50K or >50K income class. The main focus is model interpretability, not just prediction. The Streamlit app explains each prediction using the Decision Tree decision path and includes evaluation visuals such as a confusion matrix, classification report, and tree preview.

Preview

---

Project focus

• Decision Tree classification
• Explainable machine learning
• Prediction probability
• Rule-path explanation
• Confusion matrix
• Classification report
• Decision tree visualization
• Responsible evaluation and leakage awareness

Important audit note

The original legacy notebook produced a perfect score because it used a cluster feature derived from the target label. This created target leakage and made the evaluation misleading.

This demo avoids that issue by training a clean scikit-learn Pipeline using only valid input features.

Demo features

- Streamlit input form
- Predicted class output
- Prediction probability table
- Decision path explanation
- Confusion matrix
- Classification report
- Decision Tree preview
- Toggle to exclude sensitive attributes: race and sex

Live Demo

Try the Streamlit demo here:

https://decision-tree-classification-qzqfvzduowrcp5pfk5edhh.streamlit.app/

Tech stack

Python
pandas
NumPy
scikit-learn
Streamlit
matplotlib

Dataset

Dataset file expected by the app:

data/adult_dataset.csv

Target column in the original file:

Unnamed: 14

The app renames it internally to:

income_class

Target classes:

<=50K
>50K

Local setup

Create and activate a virtual environment:

python -m venv .venv

Windows:

.\.venv\Scripts\Activate.ps1

macOS/Linux:

source .venv/bin/activate

Install dependencies:

pip install -r requirements.txt

Run the app:

streamlit run app.py

Expected folder structure

.
├── app.py
├── requirements.txt
├── README.md
└── data/
    └── adult_dataset.csv

Model pipeline

The app trains a model using this workflow:

1. Load dataset
2. Replace ? values with NaN
3. Rename target column
4. Split train/test with stratification
5. Impute numeric features with median
6. Impute categorical features with most frequent value
7. One-hot encode categorical features
8. Train DecisionTreeClassifier
9. Evaluate model
10. Explain prediction using decision path

Evaluation

The app displays:

- Test accuracy
- Prediction probability
- Confusion matrix
- Classification report
- Decision Tree visualization

Responsible ML disclaimer

This app is an educational portfolio demo. It should not be used as a production decision system for employment, credit, financial eligibility, or other high-impact decisions.

The dataset includes sensitive attributes such as race and sex. The app provides an option to exclude those features from model input.