Vervolgonderzoek met synthetische data n.a.v. resultaten vanuit ARYC2025:
Can we use machine learning to identify eating disorders using unlabeled data?
I recently finished the Aletta Regional Year Challenge, where the theme this year was the "Food Environment." Using Lifelines_NL Cohort data.
We know that the environment we live in, including the pressure of social media and "perfect" body images, plays a massive role in our health. The current social media trends include toxic gym culture and the glorification of using peptides and research chemicals to make oneself appear better at the risk of unknown consequences. These trends are very dangerous and need to be watched closely.
For my project, I trained a machine learning model to identify hidden groups at risk of eating disorders within the Lifelines Cohort data.
Why is this urgent?
Eating disorders like Body Dysmorphia, BED (Binge Eating Disorder), Bulimia, and Anorexia Nervosa are often incredibly hard to spot until it is too late. They are statistically common but clinically elusive.
The Solution: Using an unsupervised machine learning model - DBSCAN.
I analyzed feature distributions to find "statistical outliers." The results were promising. I identified distinct clusters representing low-intake/low-BMI and high-intake/high-BMI groups. I also found that "noise" points in the data often corresponded to extreme behavioral outliers.
This proof of concept shows that machine learning can be a powerful tool to flag risk symptoms early, potentially allowing for earlier interventions.
For my honours diploma i will continue this research using the advice and tips I got from presenting this proof of concept. Any advice and help is welcome!
The problems to be faced now are:
- No Data from teens aged 14-18 years old, where eating disorders and body image issues are more prevalent
- Not everyone goes to the GP often, so there must be another way to collect the data for this to be a realistic working tool.
- Proper backtracing to provinces and municipalities, and what to advise them if their rate of eating disorders is significantly higher.