Counterfactual Data Augmentation for Predictive Models

FINAL REPORT DEPARTMENT OF COMPUTING IMPERIAL COLLEGE OF SCIENCE, TECHNOLOGY AND MEDICINE Counterfactual Data Augmentation for Deep Learning Predictive Models Author: Izabella Kacprzak Supervisor: Prof. Ben Glocker Co-supervisors: Dr Miguel Monteiro Dr Fabio De Sousa Ribeiro and Dr Tian Xia Second Marker: Dr Bernhard Kainz June 19, 2023 Abstract Fidelity of predictive models is often limited when they rely overly on biases in train- ing data. These biases occur due to many constraints in the pr