Improving gender classification with feature selection in forensic anthropology

Abstract

Gender classification has been one of the most vital tasks in a real world problem especially when it comes to death investigations. Developing a biological profile of an individual is a crucial step in forensic anthropology process as for the identification of gender. Forensic anthropologists employ the principle of skeleton remains to produce a biological profile. Different parts of skeleton contains different features that will contribute to gender classification. However, not all the features could contribute to gender classification and affect to a low accuracy of gender classification. Therefore, feature selection method is applied to identify the most significant features for gender classification. This paper presents the implementation of feature selection approaches which are Particle Swarm Optimization (PSO) and Harmony Search (HS) algorithm using three different dataset from Goldman Osteometric Dataset, Osteological Collection and George Murray Black Collection. All three dataset contains 4081 samples of metrics measurement and have gone through the process of classification by using Back Propagation Neural Network (BPNN) and Naïve Bayes classifier. The main scope of this paper is to identify the effect of feature selection towards gender classification. The result shows that the accuracy of gender classification for every dataset increased when feature selection is applied to the dataset. Among all the skeleton parts in this experiment, clavicle part achieved the highest increment of accuracy rate which is from 89.76% to 96.06% for PSO algorithm and 96.32% for HS.