An MAE of 2.5
Reduce the mean absolute error of age estimates to 2.5, across all Fitzpatrick skin types.
Face recognition Age verification
Age estimation with AI. Built for age verification.
A leading age verification company wanted to enhance its age estimation system. Here’s how Exometrics helped it achieve remarkable results.
1.98
Mean absolute error (MAE) for the crucial 18–24 age group
The opportunity
One clear target. An MAE of 2.5 across every skin type.
As age verification becomes crucial in more and more industries, demand for highly accurate age estimation is growing. A prominent company in the field partnered with Exometrics to develop an advanced age estimation engine, capable of reducing the mean absolute error (MAE) of its estimates to 2.5 across all Fitzpatrick skin types.
Diverse demographics
Work reliably for people from a diverse range of demographics.
Consistent results
Give consistent estimates across different ethnicities and age groups.
Fast and efficient
Run quickly and efficiently enough for real-world applications.
Our solution
Three PhDs. Three generations of engines.
The team
A highly qualified team, assembled for the challenge.
Exometrics assembled a highly qualified team to take on the challenge, made up of three PhD holders: in Applied Mathematics, Biostatistics and Computer Science.
The approach
A multi-stage approach, with a diverse set of techniques.
This expert group devised a multi-stage approach, developing three generations of age estimation engines.
Our data scientists and machine learning specialists used a diverse array of techniques, including ensemble models, deep learning architectures and cutting-edge data preprocessing.
Technologies used
The stack behind the engine.
- Google Cloud
- Docker
- Python
- PyTorch
- TensorFlow
Results
Outperforming a competing firm. In the age group that matters most.
- Developed and delivered a sophisticated age estimation system, built on state-of-the-art custom machine learning models
- Achieved an MAE of 1.98 for the crucial 18–24 age group, outperforming a competing consulting firm
- Demonstrated consistency across ethnicities, improving accuracy for under-represented groups
Mean absolute error
1.98
For the crucial 18–24 age group. Lower is better.
Need a similar solution?
Get in touch for a free consultation with our experts.
Other case studies. See how we’ve helped other clients.
850 work hours saved per month Generative AI Workflow Automation
A state-of-the-art generative AI system that automates workflows and improves efficiency.
Read the case study
95% accuracy compared with GPS Navigating using AI
A novel approach to determining location using pressure sensors and AI.
Read the case studyAdvanced Website Analytics
Finding out which visitors are most likely to convert, based on content viewed, behaviour and more.
Read the case study