A Comparative Study on Different Machine Learning Algorithms to Explore Sexual Dimorphism in Cephalometric Measurements of North Indian Population

Sex estimation from skeletal remains is an important component of personal identification in forensic anthropology. The different rates of skeletal growth and development pertaining to age, ethnicity and sex form the basis of such identification. The present study has been conducted on the contemporary North Indian Haryanvi population to ascertain the cephalometric measurements and the best machine learning algorithm to study sexual dimorphism. A cross-sectional study was conducted on 200 individuals (M: 100: F:100) aged between 18 and 40 years and 12 cephalometric measurements were obtained using spreading and sliding callipers. Statistical analysis was done using Statistical Package for Social Science (SPSS) version 21.00. All 12 variables showed sexual dimorphism and the sexing accuracy ranged between 62% and 93.5% in univariate analysis. Bizygomatic breadth and bi-gonial width (BiGoW) showed an accuracy of 99% in multivariate analysis. The receiver operating curve (ROC) analysis also depicted BiGoW to have the highest area under the curve (AUC) (1.00) and sexing accuracy of 95.5%. Principal component analysis (PCA) also revealed a similar result with BiGoW, nasal height (NH) and ZyBr having the highest communalities. However, it was concluded that discriminant function analysis (DFA) and ROC analysis showed more promising results in studying sexing accuracy as compared to PCA.