A Hybrid Approach Based on Machine Learning to Identify the Causes of Obesity

Anar TAGHIYEV, Adem Alpaslan ALTUN, Sona CAGLAR

Abstract


The obesity issue has international relevance in recent years and the study aim is to develop a hybrid classification model to identify the causes of obesity in the region of Turkey. In the period from March to November 2019, patient records retrieved from the database of Electronic Health Records (EHR) of Aksaray Sultanhani Family Health Center (ASFHC) were examined, and the questionnaire was conducted among the females aged 18 years and above. In the study, a two-stage hybrid model was used in order to better classify the collected data. The first-stage is the feature (i.e. best variable) selection while the second-stage is for classification. The performance of a proposed two-stage hybrid approach was compared with traditional single-stage classifiers: Decision Trees (DT) and Logistic Regression (LR) algorithms. In the study, the proposed hybrid system gives 91.4 of accuracy, which is better than other classifiers (i.e. 4.6 % higher than the performance of LR and 2.3 % higher than the performance of DT). Thus, the proposed hybrid system provides a more accurate classification of patients with obesity and a practical approach to estimating the factors affecting obesity.

In the future, we are going to research in detail the relationship between Type 2 Diabetes (T2D) and obesity in females.


Keywords


apache spark; machine learning; classification; hybrid intelligent systems; obesity.

Full Text: PDF