Early identification and intervention of obesity requires robust predictive models because this health problem remains a widespread global epidemic which produces multiple chronic diseases. Identifying obesity levels in individuals becomes the focus of this research through the implementation of voting-based ensemble models developed with machine learning algorithms. The first step of preprocessing data required RandomUnderSampler to equalize data class distribution. The model efficiency improved through the application of recursive feature elimination (RFE) for feature selection. The research team developed four voting-based ensemble models which combined support vector machine and decision tree (SVM-DF) and support vector machine and random forest (SVM-RF) and decision tree and random forest (DF-RF) and random forest and extra tree (RF-ET). The models use the advantages of multiple classifiers to achieve better prediction accuracy. The decision tree and random forest (DF-RF) method exceeded other models during testing because it achieved a 94% accuracy rate. Research outcomes indicate that ensemble learning produces superior obesity estimation classifications when deep forest operates with random forest. The studied hybrid analysis models demonstrate their effectiveness in medical data analysis which helps professionals make better healthcare choices.