Outlier Detection using Pseudo-Supervised Learning (PSOD)

Overview

PSOD (Pseudo-Supervised Outlier Detection) is a novel approach for detecting outliers in tabular data by treating each feature as a target variable and using the prediction errors from regression models as outlier scores.

Key Features

• 🚀 Flexible Architecture: Supports any scikit-learn compatible regressor as base learner
• 📊 Mixed Data Types: Handles both numerical and categorical features
• 🔄 Multiple Transformations: Supports logarithmic, Yeo-Johnson, and no transformation
• 🎯 Customizable Detection: Configure outlier detection on low end, high end, or both

Installation

# TODO: Add PyPI installation instructions once package is published
pip install outlier-pseudo-supervised

From Source

git clone https://github.com/diogoribeiro7/outlier_pseudo_supervised.git
cd outlier_pseudo_supervised
pip install -e .

Quick Start

from psod import PSOD
import pandas as pd

# Load your data
df = pd.DataFrame({
    'feature1': [1, 2, 3, 4, 100],
    'feature2': [10, 20, 30, 40, 1000],
    'category': ['A', 'B', 'A', 'B', 'A']
})

# Initialize PSOD
detector = PSOD(cat_columns=['category'])

# Detect outliers
outlier_scores = detector.fit_predict(df)

Documentation

TODO: Add link to full documentation once Sphinx docs are generated

Full documentation available at https://outlier-pseudo-supervised.readthedocs.io

Benchmarks

TODO: Add comprehensive benchmark results comparing PSOD with other outlier detection methods

Performance comparisons coming soon.

Contributing

TODO: Create CONTRIBUTING.md with detailed contribution guidelines

We welcome contributions! Please see our Contributing Guide for details.

Citation

TODO: Add proper citation format once paper/preprint is published

If you use this software in your research, please cite:

@software{ribeiro2024psod,
  author = {Ribeiro, Diogo},
  title = {PSOD: Pseudo-Supervised Outlier Detection},
  year = {2024},
  publisher = {GitHub},
  url = {https://github.com/diogoribeiro7/outlier_pseudo_supervised}
}

License

This project is licensed under the Apache License 2.0 - see the LICENSE file for details.

Contact

Diogo Ribeiro - @diogoribeiro7

Project Link: https://github.com/diogoribeiro7/outlier_pseudo_supervised