Examination and Evaluation of Obesity RiskFactors with Explainable Artificial Intelligence

Abstract

There is an increasing need for effective methods for the detection and management of obesity, which is an important public health problem worldwide and is critical for the sustainability of health systems. This study examines the effectiveness of data preprocessing and machine learning techniques in detecting obesity. Data preprocessing steps, including the removal of unnecessary data, handling missing values, and addressing data imbalance, are necessary to enhance the accuracy of machine learning algorithms. In this study, data preprocessing steps were applied to an obesity dataset to make it suitable for machine learning. Using a dataset of 2111 patients, this study evaluates the effectiveness of machine learning techniques in detecting obesity. Following the completion of data preprocessing, obesity was identified using various machine learning algorithms, including Decision Tree Classification, Random Forest Classification, Naive Bayes Classification, KNN, and XGBoosting, and their performances were compared. According to the results, the XGBoosting algorithm exhibited the highest accuracy (0.92), precision, recall, and F1-score values. Explainable Artificial Intelligence (XAI) techniques, such as SHAP and InterpretML, were employed to understand the effects of obesity parameters and determine which parameters have a greater impact on obesity. Through visualizing and analyzing the effects of obesity parameters, these techniques facilitated the identification of significant parameters in obesity detection. The findings demonstrate that the XGBoosting algorithm outperforms other algorithms in detecting obesity. Furthermore, XAI techniques play a crucial role in comprehending obesity parameters. Specifically, a family history of obesity and factors like FCVC and CAEC appear to have more significant effects compared to others.