Using image conversion techniques to detect adversarial examples in machine learning models for tabular data

Zahi Kapri, Edita Grolman, Ikuya Morikawa, Toshiya Shimizu, Yuval Elovici, Asaf Shabtai

Abstract: Highlights•Employed unsupervised and supervised approaches for detecting adversarial examples.•Focused on machine learning models trained on tabular data.•Converted tabular data to images, leveraging the processing capabilities of deep learning techniques.•Leveraged explainability and transfer learning to enhance adversarial detection.•Achieved a detection rate of almost 100% on most of the examined datasets.

External IDs:doi:10.1016/j.asoc.2025.113288