AutoOC: A Python module for automated multi-objective One-Class Classification

Abstract

AutoOC is an open-source Python module to efficiently automate the selection and hyperparameter tuning of quality OCC (One-Class Classification) learners. By using a GE (Grammatical Evolution) approach, AutoOC searches for five base learners, namely IF (Isolation Forest), LOF (Local Outlier Factor), OC-SVM (One-Class SVM), AE (Autoencoder), and VAE (Variational Autoencoder). The module provides a multi-objective search, where predictive performance and computational efficiency are simultaneously optimized. By providing a simple set of functions, AutoOC allows the user to easily generate OCC models for a dataset, being well-suited for anomaly detection tasks, where most of the data is composed of normal records.