Comparative Study of Obesity Levels Classification

Obesity is a growing global health problem, requiring accurate data analysis to understand and address contributing factors. The level of obesity can be identified based on eating habits and physical conditions, which consist of several parameters. However, the performance of widely used machine learning methods has not provided satisfactory results. Therefore, this study analyzes obesity data using pre-processing methods to improve data quality before classifying data. The dataset used is 2111 data and includes 17 variables/features. The classification methods are Random Forest Classifier, Light Gradient Boosting Machine (LGBM) Classifier, Decision Tree Classifier, and Extra Tree Classifier. The process of data pre-processing involves data integration, data labeling, data transformation, normalization, and data cleansing. After pre-processing the data, four algorithms were used to identify patterns in the obesity data. The Random Forest Classifier is used for its ability to handle unbalanced data and reduce the risk of overfitting. The LGBM Classifier is used for a probabilistic approach to classification. The Decision Tree Classifier is applied for straightforward interpretation and clear understanding of patterns, while the Extra Tree Classifier is applied to improve the variety and accuracy of classification. The experimental results showed that a good data pre-processing method significantly improved the performance of the classification. Among the four algorithms tested, the Random Forest Classifier and Extra Tree Classifier performed best in accuracy and generalizability. Combining appropriate data pre-processing with powerful classification algorithms can provide deep insights to address obesity problems and formulate effective public health interventions.