Machine learning framework for predicting susceptibility to obesity

Abstract Obesity, currently the fifth leading cause of death worldwide, has seen a significant increase in prevalence over the past four decades. Timely identification of obesity risk facilitates proactive measures against associated factors. In this paper, we proposed a new machine learning framework for predicting susceptibility to obesity called ObeRisk. The proposed model consists of three main parts, preprocessing stage (PS), feature stage (FS), and obesity risk prediction (OPR). In PS, the used dataset was preprocessed through several processes; filling null values, feature encoding, removing outliers, and normalization. Then, the preprocessed data passed to FS where the most useful features were selected. In this paper, we introduced a new feature selection methodology called entropy-controlled quantum Bat algorithm (EC-QBA), which incorporated two variations to the traditional Bat algorithm (BA): (i) control BA parameters using Shannon entropy and (ii) update BA positions in local search using quantum mechanisms. Then, these selected features fed into several machine learning (ML) algorithms, including LR, LGBM, XGB, AdaBoost, MLP, KNN, and SVM. The final decision was obtained based on the majority voting. Experiment results demonstrated that the proposed EC-QBA outperformed the most recent feature selection methodology in terms of accuracy, precision, sensitivity, and F-measure. It introduced 96% accuracy, 96% precision, 96.5% sensitivity, and 96.25% F-measure. Additionally, experimental results indicated that the EC-QBA with the proposed OPR model delivered the best performance, surpassing modern strategies for predicting obesity by achieving maximum accuracy.