3 Researchers develop indigenous machine for forensic age estimation

Three Ghanaian researchers have achieved a major scientific breakthrough with the development of the country’s first indigenous machine learning model for forensic age estimation among adolescents.

The pioneering research titled “Prediction of Radiological Forensic Age of Adolescents in Ghana Using Machine Learning Algorithms” was conducted in the Ho Municipality and is expected to improve age verification practices in the country significantly.

It will also provide Ghana with a more accurate, reliable and culturally relevant tool for forensic age estimation. Experts believe the model could support fairer and more reliable decision-making in juvenile justice systems, immigration and asylum procedures, child protection services, adoption processes and sports age verification, particularly in football.

The use of non-invasive ultrasound imaging combined with low-radiation dental imaging also makes the approach safer and more practical for wider application.

The research was conducted by Dr Frempong Acheampong, the principal researcher, with Professor Eric Kwasi Ofori and Professor Seth Owusu-Agyei as supervisors.

For decades, Ghana has relied heavily on foreign forensic age estimation standards, especially the Greulich-Pyle Atlas, which was developed in the 1950s using American Caucasian children.

Experts say these standards often produce inaccurate results when applied to African populations due to differences in genetics, nutrition, socio-economic conditions and environmental influences.

These inaccuracies have created challenges in the juvenile justice administration, immigration and asylum processes, child protection services, school placements and sports age verification, where minors may mistakenly be treated as adults or adults falsely identified as children.

The study

Speaking in an interview with the Daily Graphic, Dr Acheampong said that to address these challenges, they conducted a population-specific study involving 265 healthy adolescents aged between 14 and 19 years in a senior high school in the Ho Municipality.

The study, he said, collected socio-demographic information, anthropometric measurements including Body Mass Index (BMI), bilateral knee ultrasound scans and dental radiographs of the first three molars.

Using R software, he said, they developed and tested four machine learning algorithms: Random Forest, Generalised Linear Model (GLM), Recursive Partitioning and Regression Trees (RPART), and K-Nearest Neighbours (KNN) 

Key findings

Dr Acheampong said the findings revealed that the Random Forest model achieved the highest classification accuracy, with measurements from the right fibula region identified as the strongest skeletal predictor of age.

The study, he added, also found that the pulp-to-root width ratio at the cementoenamel junction of the second molar was the strongest dental predictor.

In addition, he said the BMI emerged as an important contributor to age estimation, confirming the influence of nutritional status on biological maturation.

He said that although the Random Forest model achieved the highest classification accuracy, the Generalised Linear Model (GLM) demonstrated the most balanced overall performance across key indicators, including sensitivity, specificity, accuracy, F1-score and Area Under the Curve (AUC).

He said they further reported a strong positive correlation between the model’s predicted age and the actual chronological age.

Major step for African science

Dr Acheampong described the research as a landmark contribution toward the development of African-centred scientific and technological solutions.

He said that beyond Ghana, the study could serve as a model for other countries across sub-Saharan Africa seeking more accurate and locally relevant forensic age estimation systems.