Serviços Personalizados
Journal
Artigo
Indicadores
Citado por SciELO
Acessos
Links relacionados
Citado por Google
Similares em SciELO
Similares em Google
Compartilhar
Revista EIA
versão impressa ISSN 1794-1237versão On-line ISSN 2463-0950
Resumo
BECERRA-ALVAREZ, Johana Gabriela e CORTES-OSORIO, Jimy Alexánder. Supervised Machine Learning Proposal For The Estimation Of Human Biological Age Based On Forensic Dentistry. Rev.EIA.Esc.Ing.Antioq [online]. 2020, vol.17, n.34, pp.240-258. Epub 05-Set-2021. ISSN 1794-1237. https://doi.org/10.24050/reia.v17i34.1392.
Colombian Criminal Procedure Code supports the appropriate use of dental pieces for the identification of individuals, so this technique is a tool of great importance for the administration of justice institutions. The classic methods of human identification, especially those used to estimate the biological age of death in adults such as Lamendin, have been developed using information from populations with morphological, metric and cultural characteristics different from those in Colombia. Consequently, Colombian forensic specialists have obtained high error rates in age estimation results. In this research, the biological age was initially calculated using the classic method of Lamendin. Subsequently, with the direct and indirect measures, also used in the Lamendin method, the training of machine learning techniques was carried out knowing the chronological age of the individuals in the sample. The results of the supervised machine learning regression techniques were compared: Vector Support Machines (SVM), Gaussian Regression Processes (GPR) and Ensembles of Trees (EoT), among others. A sample of 48 single-root dental pieces belonging to 45 individuals of Colombian nationality was studied, with chronological ages of death known within the age range from 19 to 81 years.
Additionally, the digital caliper gauge and the comparison microscope were used to measure the heights of the morphological characteristics in the dental pieces. The best accuracy in the estimation of the biological age was achieved with the GPR regression that showed a root mean square error (RMSE) of 3.37 years in the identification, which was compared with the accuracy reached by the Lamendin method with a RMSE of 15.52 years. This research shows that the GPR machine learning regression, especially for the Colombian case, is a valid tool for estimating the biological age of death with much lower errors than those achieved by traditional European techniques.
Palavras-chave : biological age; forensic dentistry; human identification; Lamendin; Machine learning; tooth.