Age-at-death estimation through neural networks: an evaluation of DRNNAGE software for age-at-death estimation on a Dutch medieval skeletal sample

Exploring the junction of artificial intelligence (AI), osteology and, forensic anthropology, this thesis validates the application of neural networks (NN) for accurate age-at-death estimation in skeletal remains. Because of the lack of accurate age-at-death estimation methods and the discrepancy between biological and chronological ages, there a high demand for objective and unbiased approach. A previous study developed DRNNAGE, a NN solution to estimate age-at-death and reported a promising accuracy in predicting age-at-death (95%) and a mean absolute error (MAE) of ~6 years. In this study, the reproducibility and the accuracy of DRNNAGE prediction will be validated employing an archaeological Dutch medieval skeletal sample from the Middenbeemster collection (MB11). The sample consisted of 52 individuals with an age range of 19-101 years. Through a multifactorial transition analysis, 101 features were scored according to two or three levels of senescent change. The results show...Show moreExploring the junction of artificial intelligence (AI), osteology and, forensic anthropology, this thesis validates the application of neural networks (NN) for accurate age-at-death estimation in skeletal remains. Because of the lack of accurate age-at-death estimation methods and the discrepancy between biological and chronological ages, there a high demand for objective and unbiased approach. A previous study developed DRNNAGE, a NN solution to estimate age-at-death and reported a promising accuracy in predicting age-at-death (95%) and a mean absolute error (MAE) of ~6 years. In this study, the reproducibility and the accuracy of DRNNAGE prediction will be validated employing an archaeological Dutch medieval skeletal sample from the Middenbeemster collection (MB11). The sample consisted of 52 individuals with an age range of 19-101 years. Through a multifactorial transition analysis, 101 features were scored according to two or three levels of senescent change. The results show that the DRNNAGE provides a considerably reliable estimate of age-at-death with an accuracy of 87.9%, with a relatively strong correlation between the estimated and archival ages (Spearman’s r=0.7204). Interestingly, DRNNAGE performed with improved accuracy on individuals over 50 years. In conclusion, DRNNAGE is recommended for applications in elderly individuals and is suitable in archaeological contexts. Further research into different population contexts is needed. As the implementation of AI is still in the early stages, the possibilities of AI collaboration can achieve are infinite.Show less

Leiden University Student Repository

Documents

In Collections

Age-at-death estimation through neural networks: an evaluation of DRNNAGE software for age-at-death estimation on a Dutch medieval skeletal sample