Penanganan Imbalance Data pada Hasil Klaster dengan SMOTE untuk Prediksi Permintaan Perusahaan Ekspedisi Menggunakan XGBoost

Abstract

The rapid growth of online shopping has driven the increasing need for accurate demand prediction in logistics and courier service companies. However, this demand presents challenges due to imbalanced data, which causes predictive models to be biased toward the majority class. This study proposes a combined approach using the Synthetic Minority Over-sampling Technique (SMOTE) and K-Means clustering to address data imbalance, along with the Extreme Gradient Boosting (XGBoost) algorithm as the predictive model. A historical dataset consisting of 45,684 entries was used, including features such as quantity, unit, weight, and destination. The research stages included preprocessing, normalization, clustering, evaluation (using silhouette score, Davies-Bouldin index, and Calinski-Harabasz score), and oversampling of minority clusters. The application of SMOTE for handling imbalanced data proved to enhance model performance, Despite the enhancement being rather modest owing to the initial model's already robust performance. Nevertheless, in the context of imbalanced data, such improvement is meaningful as it indicates that the minority class receives more balanced attention from the model.