An optimized ensemble learning model for interpretable and efficient obesity classification.

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Computational Biology

Background:

• Obesity is a complex health issue influenced by genetic, behavioral, and environmental factors.
• Existing machine learning and statistical methods for obesity classification show promise but require further improvement.
• Accurate obesity classification is crucial for effective diagnosis and personalized treatment strategies.

Purpose of the Study:

• To propose and evaluate an optimized stacking machine learning classifier (SMLC) for enhanced obesity classification.
• To investigate the effectiveness of combining diverse machine learning and deep learning models through an optimized stacking ensemble.
• To assess the interpretability and clinical relevance of the proposed SMLC model for decision support.

Main Methods:

• Developed a two-layer stacking machine learning classifier (SMLC) integrating seven base algorithms: random forest, gradient boosting, deep learning, support vector machine, k-nearest neighbor, light gradient-boosting machine, and extreme gradient boosting.
• Employed Bayesian optimization to fine-tune the parameters of the base models.
• Utilized logistic regression in the second layer to aggregate base model outputs for final obesity classification.
• Employed SHAP and other interpretation models to analyze model results and clinical relevance.

Main Results:

• The proposed SMLC model achieved superior performance in obesity classification compared to individual base models and existing counterpart methods on a benchmark dataset of 2111 instances.
• The ensemble approach effectively leveraged the strengths of diverse algorithms, leading to improved classification accuracy and robustness.
• Interpretability analysis confirmed the clinical relevance and predictive reliability of the SMLC model.

Conclusions:

• The optimized stacking machine learning classifier (SMLC) presents a competitive and interpretable decision support system for obesity classification.
• The SMLC model has the potential to assist physicians in clinical practice, pending validation on external cohorts.
• Further research into ensemble methods and interpretable AI is warranted for advancing obesity management.