A Multi-class Obesity Risk Prediction Using Machine Learning and Explainable Artificial Intelligence

Because of the intricate relationships between dietary practices, physical characteristics, and lifestyle, weight-related health issues have grown to be a global concern. The shortcomings of current machine learning (ML) techniques include inefficient feature selection, imbalanced datasets, a binary classification focus, decreased accuracy, and inadequate hyperparameter tweaking. This paper uses a clinically validated dataset of 1,638 patients from Bangladeshi healthcare facilities to provide a complete framework for ML-based multiclass obesity risk prediction in order to fill these gaps. The proposed approach combines 5-fold cross-validation with GridSearchCV for systematic hyperparameter tuning and ensemble feature selection. The Synthetic Minority Over-sampling Technique was used just on the training set to address class imbalance and guarantee balanced learning across the seven weight categories. The proposed XGBoost outperformed the other ML algorithms that were assessed for obesity risk prediction due to their high accuracy score of 95.4%. According to the results, the proposed method outperformed previous works by at least 5.18% in accuracy increase and 10.30% in F1-score ga