Sex inference based on convolutional neural network analysis of fingerprint data

Area of Science:

• Forensic Science
• Biometrics
• Artificial Intelligence

Background:

• Determining biological sex from fingerprints is an emerging forensic science challenge.
• Traditional methods for sex determination from fingerprints have limitations.
• Convolutional Neural Networks (CNNs) offer a potential automated approach.

Purpose of the Study:

• To develop and validate a lightweight CNN for inferring biological sex from fingerprint features.
• To evaluate the CNN's accuracy and generalisation ability on independent datasets.
• To enhance the interpretability of the CNN model using Class Activation Mapping (CAM).

Main Methods:

• A dataset of 1,000 fingerprint images from 200 volunteers was curated.
• A dual-convolutional-layer CNN was designed and optimised.
• The model was evaluated using an independent test set and fivefold cross-validation.
• Class Activation Mapping (CAM) was used for visualizing model focus.

Main Results:

• The CNN achieved a validation accuracy of 91.00% and a test accuracy of 95.00%.
• Area Under the Curve (AUC) values reached 0.974 (validation) and 0.983 (test).
• Fivefold cross-validation confirmed stable performance with a mean accuracy of 90.60% (SD: 2.04%).

Conclusions:

• The developed CNN is an efficient and reliable tool for inferring biological sex from fingerprints.
• The model demonstrates comparable or superior performance to traditional forensic methods.
• The CNN shows potential as a complementary tool in forensic identification, with enhanced interpretability through CAM.