Obesity is no longer just a public-health problem; it’s a silent killer. Ranked as the 5th leading global cause of death, its prevalence has surged over the past forty years, dragging along higher rates of type 2 diabetes, heart disease and more. What if we could spot those at risk before the weight gain snowballs into disease?
That’s the findings ehind the new systematic review on AI-enabled obesity prediction published in International Journal of Medical Informatics. The authors reviewed through 6,351 papers across major bibliographic databases – PubMed, Scopus, and Web of Science, focusing on cohort studies up to March 2024. Their goal was to figure out where AI stands in spotting obesity risk early, and what’s holding it back.
Key Findings by the Authors:
The cohorts in these studies varies in sample size and durations
Most researchers gravitate to supervised learning by using variousb ML algorithms such as random forest, linear regression, support vector machines, and logistic regression.
The most used inputs? Demographics (age, gender, socioeconomic factors) and biomarkers (blood tests, genetic markers, metabolic indicators).
When it comes to performance, artificial neural networks (ANNs) performed better
Supervised ML means teaching computers with examples so they can learn to predict answers or sort things correctly.
Deep learning models like Artificial Neural Network (ANN) exhibits superior performance in terms of AUC (Area Under the Curve) and k-means in terms of accuracy. The review although highlights gaps like inconsistent data quality, limited external validation, and few studies pushing into more advanced AI (deep learning, ensemble methods, explainable AI). What this really means is that although AI shows promise, we can’t yet rely on its predictions in clinics or public-health settings without more rigorous work. But there’s a catch: it’s early days.
For dietitians, researchers, and clinicians building predictive models or AI decision support tools, this review is a kind of roadmap. It pinpoints where the field is now and what’s needed next like standardizing cohorts, sharing data, exploring newer algorithms, and validating in real world settings.
