A system and method is disclosed for improving fraud detection in the context of a user submitting, via a client device, a photo of a photo ID and a selfie taken during a step of the verification process. A machine learning model may be trained to generate biometric data signals from the selfie. As examples, the biometric data signals generated from the selfie can include an estimated age, gender, hair color, and eye color. The biometric data generated from the selfie may be compared with biometric data listed on the photo ID as an aid to identifying potential identity fraud. In some implementations, a facial filter corresponding to a map of a set of facial measurements of the photo in the photo ID may be compared with the facial measurements of the selfie.