An ensemble-based feature selection framework to select risk factors of childhood obesity for policy decision making

• DOI: 10.1186/s12911-021-01580-0
• Publication Date: 2021-07-21T04:02:58Z
• Authors: Xi Shi; Gorana Nikolic; Gorka Epelde; Mónica Arrúe; Joseba Bidaurraza...; Roberto Bilbao; Bart De Moor
• Keywords: Interpretability; Discriminative model; Feature (linguistics); Lasso; Ensemble Learning

Abstract Background The increasing prevalence of childhood obesity makes it essential to study the risk factors with a sample representative population covering more health topics for better preventive policies and interventions. It is aimed develop an ensemble feature selection framework large-scale data identify good interpretability clinical relevance. Methods We analyzed collected from 426,813 children under 18 during 2000–2019. A BMI above 90th percentile same age gender was defined as overweight. An framework, Bagging-based Feature Selection integrating MapReduce (BFSMR), proposed factors. comprises 5 models (filter mutual information/SVM-RFE/Lasso/Ridge/Random Forest) filter, wrapper, embedded methods. Each model identified 10 variables based on variable importance. Considering accuracy, F-score, characteristics, were classified into 3 levels different weights: Lasso/Ridge, Filter/SVM-RFE, Random Forest. voting strategy applied aggregate selected features, both weights taken consideration. compared our another two selecting top-ranked features in terms 6 dimensions interpretability. Results Our method performed best select top by BFSMR are age, sex, birth year, breastfeeding type, smoking habit diet-related knowledge mothers, exercise, Mother’s systolic blood pressure. Conclusion provides solution identifying diverse interpretable set without bias data, which can help potentially some other diseases future interventions or policies.