XG Boost Based Data Imputation and Outlier Detection Methods for Classification of Stunting

Since stunting was recognized as a global issue by the WHO, numerous countries have conducted research on this topic. Stunting classification heavily depends on the quality of the datasets used. The Demographic Health Survey (DHS) dataset is commonly employed in such research but often contains missing and outlier data, which negatively impacts classification accuracy. This study proposes data imputation and outlier detection methods at the pre-processing stage to enhance stunting classification accuracy. Six imputation methods Max, Min, Mean, Mode, Median, and KNN Imputation—are combined with three outlier detection methods: Boxplot, Z-score, and Interquartile Range (IQR). XGBoost, a machine learning algorithm known for its robustness in handling diverse data types, is used for the classification task. Experimental results demonstrate that pre-processing the data through imputation and outlier detection improves stunting classification accuracy, achieving rates above 95%. The highest accuracy was obtained using median imputation.