Abstract

Sex determination constitutes one of the most critical stages in the identification process within forensic anthropology. The morphological characteristics of the hand such as shape, size, finger lengths, and the proportional relationships among these measurements exhibit distinct differences between sexes and therefore serve as important biometric indicators in sex estimation. This study aimed to determine individuals’ sex from hand photographs using a deep feature–based machine learning approach. A total of 400 healthy volunteers (192 males and 208 females) without any pathology, deformity, or history of surgical intervention in the hand region were included in the study. Hand photographs were obtained using a professional camera and tripod under standardized conditions with a fixed angle and distance and homogeneous lighting. A hybrid approach was used in which anonymized images were processed with deep learning for feature extraction and evaluated by machine learning classifiers. The highest accuracy in sex estimation was achieved with the Quadratic SVM (Support Vector Machine) model (training accuracy: 95.92%; test accuracy: 95.96%). In addition, the Cubic SVM and Medium Gaussian SVM models also demonstrated comparably high classification performance. The analysis of hand images using deep learning–based feature extraction combined with machine learning classifiers yielded high levels of accuracy and reliability in sex prediction. The findings demonstrate that proposed models provide highly accurate and reliable results for sex estimation from hand photographs. The effectiveness and reproducibility of these models offer a powerful alternative to traditional anthropometric methods. Accordingly, this study is expected to make significant scientific and methodological contributions to the fields of anatomy, forensic medicine, and forensic anthropology.

Key words: Hand, sex estimation, deep learning, forensic anthropology