Efficient Machine Unlearning Methods for Incremental Data Deletion in Supervised Learning

Abstract

Machine unlearning aims to efficiently remove the influence of specific training data from machine learning models without full retraining. This capability is crucial for privacy, data ownership, and compliance with regulations such as GDPR. In this paper, we study three representative approaches for machine unlearning in supervised learning: exact retraining, approximate unlearning using influ- ence functions, and selective retraining on affected samples. We perform toy experiments on classical sklearn datasets (Iris and Breast Cancer) to empirically evaluate the accuracy and computational trade- offs of these methods. The results illustrate key performance differences and practical considerations for deploying machine unlearning techniques.