PassGuard Password Strength Classifier

PassGuard is a machine learning notebook project for classifying password strength from text patterns. It uses a labeled password dataset, character-level TF-IDF features, and compares Logistic Regression, AdaBoost, and XGBoost models.

Project Contents

Path	Description
main.ipynb	Data loading, preprocessing, visualization, model training, cross-validation, and evaluation.
data.csv	Password dataset with password and strength columns.
verify_project.py	Lightweight reproducibility check for the dataset/vectorizer/model pipeline.
notebook-requirements.txt	Python dependencies for running the notebook and verification script.
index.html, style.css, script.js	Static public showcase page for GitHub/Vercel.

Dataset

The notebook loads data.csv with pd.read_csv("data.csv", on_bad_lines="skip").

• Loaded rows: 669,640
• Clean rows after dropping one null password: 669,639
• Labels: 0 weak, 1 medium, 2 strong
• Class counts:
  • 0: 89,701
  • 1: 496,801
  • 2: 83,137

Approach

1. Load and clean the password dataset.
2. Tokenize each password at the character level.
3. Convert passwords into TF-IDF feature vectors.
4. Split the dataset into training and test sets.
5. Compare Logistic Regression, AdaBoost, and XGBoost.
6. Evaluate with accuracy, precision, recall, F1 score, cross-validation, and confusion matrices.

Recorded Notebook Results

Model	Holdout Accuracy	Weighted Precision	Weighted Recall	Weighted F1
Logistic Regression	0.8197	0.8028	0.8197	0.8026
AdaBoost	0.8320	0.8314	0.8320	0.7911
XGBoost	0.9805	0.9804	0.9805	0.9804

The strongest recorded model in the notebook is XGBoost, with about 98.05% holdout accuracy.

Course Context

This project was completed as part of the Machine Learning course at UNICA, Università degli Studi di Cagliari.

Run Locally

python3 -m venv .venv
source .venv/bin/activate
pip install -r notebook-requirements.txt

Run the quick verification script:

python verify_project.py

Run on the full dataset:

python verify_project.py --full

Open the notebook:

jupyter notebook main.ipynb

Showcase

The static showcase page is served from index.html, so Vercel can use the repository root as the project root.

For local preview:

python3 -m http.server 8000

Then open http://localhost:8000.