Segmentasi Pelanggan E-Commerce Berdasarkan Pola Pembatalan dan Pengembalian Pesanan Menggunakan K-Means

Abstract

Penelitian ini mengkaji segmentasi pelanggan e-commerce berdasarkan perilaku kritis pembatalan (cancellation) dan pengembalian (return) menggunakan algoritma K-Means sebagai proof-of-concept. Menggunakan Pakistan E-Commerce Dataset tahun 2017, penelitian melakukan pra-pemrosesan, rekayasa fitur perilaku (Cancellation Rate, Return Rate, Average Order Value, Discount Sensitivity, Preferred Payment Method, dan Total Orders), normalisasi Min–Max, dan pemodelan K-Means. Validasi jumlah klaster dilakukan dengan Elbow Method, Silhouette Score, serta verifikasi visual menggunakan PCA. Hasil menunjukkan K = 3 klaster yang stabil, Price-Sensitive Customers (69.48%), transaksi bernilai tinggi namun sensitif harga, Loyal Customers (13.55%), frekuensi dan konsistensi pembelian tinggi dengan CR/RR rendah, dan High-Risk Customers (16.97%), tingkat pengembalian tinggi dengan kontribusi nilai rendah. Temuan membuktikan bahwa K-Means efektif untuk mengidentifikasi pola perilaku terkait cancellation dan return dan menyediakan dasar bagi rekomendasi konseptual untuk manajemen risiko dan analisis lanjutan.This study examines e-commerce customer segmentation based on cancellation and return behaviors using K-Means clustering as a proof-of-concept. Using the Pakistan E-Commerce Dataset (2017), we performed preprocessing, behavioral feature engineering (Cancellation Rate, Return Rate, Average Order Value, Discount Sensitivity, Preferred Payment Method, and Total Orders), Min–Max normalization, and K-Means modeling. Cluster number validation relied on the Elbow Method, Silhouette Score, and PCA visualization. Results indicate K = 3 stable clusters: Price-Sensitive Customers (69.48%) high per-order value but price-sensitive; Loyal Customers (13.55%), high frequency and low CR/RR; and High-Risk Customers (16.97%), high return rate with low value contribution. The findings demonstrate K-Means’ effectiveness in identifying cancellation/return patterns and provide a conceptual basis for risk management and further analysis.