Back-Propagation neural network for gender determination in forensic anthropology

Abstract

Determination of gender is the foremost and important step of forensic anthropology in determining a positive identification from unidentified skeletal remains. Gender determination is the classification of an individual into one of two groups, male or female. The classification technique most used by anthropologists or researchers is traditional gender determination with applied linear approach, such as Discriminant Function Analysis (DFA). This paper proposed non-linear approach specific Back-Propagation Neural Network (BPNN) to determine gender from sacrum bone. Sacrum bone is one part of the body that is usually regarded as the most reliable indicator of sex. The data used in the experiment were taken from previous research, a total of 91 sacrum bones consisting of 34 females and 57 males. Method of measurement used is metric method which is measured based on six variables; real height, anterior length, anterior superior breadth, mid-ventral breadth, anterior posterior diameter of the base, and max-transverse diameter of the base. The objective of this paper is to examine and compare the degree of accuracy between previous research (DFA) and BPNN. There are two architectures of BPNN built for this case, namely [6; 6; 2] and [6; 12; 2]. The best average accuracy obtained by BPNN is model [6; 12; 2] with accuracy 99.030 % for training and 97.379 % for testing on experiment lr = 0.5 and mc = 0.9, then obtained Mean Squared Error (MSE) training is 0.01 and MSE testing is 1.660. Previous research using DFA only obtained accuracy as high as 87 %. Hence, it can be concluded that BPNN provide classification accuracy higher than DFA for gender determination in forensic anthropology.