Forensic Gender Discrimination in Malaysian Population Using Machine Learning Methods
Loong Chuen Lee, Muhammad Yusuf Adam, Siti Norfaraan Abd Sanih, Nur Izzati Bohari · 2020
Latent fingerprint is one of the most encountered evidence in a crime scene and is useful for personal identification. In forensic investigation, reliability of the identification process is greatly affected by the visibility of the minutiae features. This is because the matching of known and unknown fingerprints achieved according to the types and locations of the minutiae features. However, most of the fingerprints recovered from the crime scene are of low quality, i.e. incomplete print or minutiae features. Under such circumstances, forensic analyst can try to determine gender of donor of the latent fingerprint. This piece of information could help to narrow down the scope of searching of suspect. Forensic discrimination of gender based on ridge counts has been proposed for a few decades ago. Despite multiple works have reported the use of machine learning methods in gender classification using fingerprints, the fingerprint features were mainly extracted from the images. In this work, the diagonal ridge counts were calculated manually within a well-defined region, i.e. 25 centimetres squared. This approach is more relevant to a real crime scene investigation. This work employed two well-known machine learning algorithms, i.e. naïve Bayes (NB) and Classification and Regression Trees (CART) algorithms, in discriminating gender based on the ridge counts. The performances of predictive models have been assessed by ethnicity and finger digit via bootstrapping without replacement approach. Results showed one-digit samples can perform as good as the five-digit or ten-digit samples. Comparing to the global predictive model, ethnicity-specific models of Indian and Malay subjects, respectively, showed better improvement. Moreover, by considering all five digits of a particular hand as input data, NB tended to outperform CART; whilst the relative performances reversed when only one digit was considered as input data. In conclusion, fingerprint ridge counts can be a potential indicator of gender in the Malaysian population.