Determination of Sex and Race in Forensic Anthropology: A Comparison of Artificial Neural Network and Support Vector Machine

One of the topics covered in forensic anthropology is an investigation of skeletal remains where various properties of the skeleton are to be determined. Typically, the sample found is incomplete, meaning some bone parts are missing or destroyed, and the analysis needs to depend on limited information obtained from what is available. This research focuses on arm, leg, clavicle, and scapula bones, with 8 bone parts in total. Each part is either used independently from the other or considered altogether (aggregate) to test its usability in finding out the owner’s identity when facing such a situation. Bone measurements obtained from the database were used as input data for two different classifiers, namely artificial neural networks and supporting vector machines, with two identification targets, namely sex and race. All of the input data came from publicly available Robert J. Terry Anatomical Skeletal Collection Postcranial Osteo-metric database. Accuracies of 86.67% and 70.78% are obtained for those targets using clavicle and aggregate, respectively, showing that using all information possible from the sample rather than focusing on a single bone part is sometimes useful in improving identification accuracy.