Amrant Singh* , Savid Gershony
Background: Palatal rugae are a series of ridges on the hard palate of the mouth with high durability and stability, ensuring their usefulness as a tool for disaster victim identification. Additionally, certain characteristics of palatal rugae are shared by gender or within specific age groups. Currently, forensics odontologists must manually examine palatal rugae imprints to determine an individual’s gender, a time-consuming process vulnerable to bias and human error.
Methodology: This project sought to automate the process of palatoscopy based classification by developing the Rugae Classification Sequence (RCS), a comprehensive tool for gender identification based on features of the palatal rugae. First, data important to palatal rugae–namely rugae length, width, subject age, and gender–were extracted from anonymized images of the palate to provide a bed of training and testing data. The dataset was fed into a series of machine learning algorithms, specifically random decision forest, decision tree classifier, logistic binary classifier, and K-nearest neighbors. Each model then underwent extensive hyperparameter tuning to maximize accuracy and robustness in predicting gender given anatomical properties of the palatal rugae.
Results: On the testing set, the K-nearest neighbor’s algorithm achieved the highest accuracy score and specificity at 65% and 68% respectively, satisfying design requirements of a minimum 60% accuracy in gender classification.
Conclusion: The RCS is a novel application of machine learning in palatoscopy, and with these results, has the potential for large-scale application-indeed, it may provide forensics experts with a more efficient and reliable tool to identify victims given dental remains.