Bias-reduced dental age estimation in children: Machine-learning evaluation and a novel constrained random forest adaptation of the Cameriere method
Forensic Science International, cilt.385, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 385
- Basım Tarihi: 2026
- Doi Numarası: 10.1016/j.forsciint.2026.112979
- Dergi Adı: Forensic Science International
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, BIOSIS, CINAHL, Criminal Justice Abstracts, EMBASE, MEDLINE
- Anahtar Kelimeler: Cameriere, Dental age, Forensic anthropology, Machine learning, Systematic bias
- İnönü Üniversitesi Adresli: Evet
Özet
Dental age estimation plays a critical role in clinical and forensic applications. Because teeth are highly resistant to environmental degradation, dental-based methods remain valuable when skeletal structures deteriorate. Although integrating ML algorithms into traditional age estimation techniques has improved accuracy, classical linear regression and commonly used ML regressors often exhibit systematic bias, a phenomenon not sufficiently examined in dental age estimation. This study compared the performance of the Cameriere method and frequently used ML algorithms in Turkish children and introduced a modified Random Forest approach designed to reduce systematic bias. A total of 1190 orthopantomographs from children aged 5–15 years (593 girls, 597 boys) were evaluated. Linear Regression, Decision Tree, Random Forest, Support Vector Regression, Multilayer Perceptron, Gradient Boosting, and Extreme Gradient Boosting were compared with the Cameriere formula and the proposed modified Random Forest models. Performance was assessed using Mean Absolute Error, Root Mean Squared Error, the slope of residuals relative to chronological age, and age-group–specific error distributions. All ML methods showed higher accuracy than the Cameriere formula, with similar performance levels across ML models. The modified Random Forest models achieved the lowest Mean Absolute Error values (0.583–0.585 years). Classical ML models displayed significant systematic bias, whereas this pattern was not observed in the modified approaches.These findings indicate that, in age estimation research, evaluating error distribution alongside overall accuracy is of critical forensic importance. Future studies should focus on developing methods that further improve accuracy while rigorously minimizing systematic bias.