Gender classification using a PSO-based feature selection and optimised BPNN in forensic anthropology

Abstract

Gender classification is a crucial task in most forensic cases.In most cases, skeleton remains are employed and there are different parts of human skeleton available for the classification process.Every part of skeleton contains different types of features which benefits toward gender classification.However, some features cannot contribute toward classification as features carry no information on gender.Hence, this article proposed a particle swarm optimisation-based (PSO) feature selection and optimised BPNN model as a gender classification framework.Initially, PSO selects the most significant features that lead to an accurate classification process.In the BPNN process, the parameter tuning based on cross-validation technique is applied where the model is able to find a good combination of learning rate and momentum.This article utilised data from Goldman Osteometric dataset, Clavicle collection, and George Murray Black collection.The result shows that the accuracy of gender classification is improved for every dataset via the proposed framework.