Feasibility Study on Optimising the Efficacy of a Population Age Estimation Model for South China by Combined Machine Learning for the Second and Third Molars.

Dental age estimation, as an important part of forensic anthropology, has a wide range of applications for its results in legal practice. Given the lowered legal age for criminal responsibility in China and the increasing juvenile delinquency, we establish a morphological database targeting the second (M2) and third molars (M3) of the Southern Chinese population. Full mouth orthopantomography from 1486 individuals aged 8.00 to 24.99 years were collected and categorized into five age groups, comprising four age nodes: 12, 14, 16 and 18 years. The Demirjian method assesses M2 and M3 development, and stepwise regression analysis confirms M2's suitability for age estimation. Advanced ML algorithms, such as Random Forest (RF) and Support Vector Machine, are implemented to fit a classification model, evaluated by accuracy. Ultimately, we constructed age estimation models employing techniques such as Decision Trees, AdaBoost, and Voting methods, and assessed their performance using metrics like the mean absolute error (MAE). Among the age estimation models based on different age groups, the Voting model exhibited the most optimal performance, with an average MAE of 0.7207. The estimation model for the 12-14 age group has the highest accuracy, with an average MAE of 0.5081. The RF model has the highest accuracy in the age estimation model for the 12-14 age group, with an MAE of 0.4248. This study effectively integrates multiple ML algorithms to enhance the precision of dental age estimation using M2 and M3, providing a robust method and predictive scheme for forensic practices in ascertaining the age of criminal responsibility.