An Interpretable Machine Learning Approach for Identification of the Risk Factors of Early-Stage Overweight and Obesity

The increasing overweight and obesity rates pose a significant risk to public health globally, as it is closely associated with various diseases and higher rates of morbidity and mortality. Prompt and effective intervention in the early stages is therefore crucial. Furthermore, identifying obesity risk factors is essential to combat the growing rates of several secondary health problems such as cardiovascular disease, type II diabetes, and other comorbidities. This study aims to unveil the key factors playing a remarkable role in shaping this situation, utilizing the immense potential of machine learning methods. The proposed pipeline implements feature selection using Recursive Feature Elimination to effectively reduce the dimensionality of the dataset. The extracted subset of the original secondary dataset is then employed to train different state-of-the-art machine learning classifiers, including Logistic Regression, Random Forest, Decision Trees, along with the proposed novel stacking ensemble model. The proposed model outperforms other models, achieving a noteworthy accuracy of 99.58% with an AUC score of 1.0, when combined with the SMOTE-ENN balancing technique. Additionally, model explainable tool SHAP is used to rank the features according to their individual contribution, providing a more in-depth explanation of how each feature impacts the model outcome. By leveraging advanced machine learning algorithms and extensive datasets, this approach can significantly improve early-stage obesity prediction by uncovering its underlying causes.