Prediksi Churn Pelanggan B2B dengan Segmentasi Menggunakan Bisecting K-Means dan Long Short-Term Memory

Wardina, Khairina (Magister Ilmu Komputer University Budiluhur)
Rahmat, Ade (Universitas Budi Luhur)
Syafrullah, Mohammad (Universitas Budi Luhur)

Faktor Exacta
Vol 18, No 3 (2025)

In the competitive B2B sector, customer churn is a key challenge, particularly for a Fast-Moving Consumer Goods (FMCG) distribution company in North Sulawesi. From 2019 to 2022, the company's churn rate continued to rise despite efforts to reduce it. This study addresses the challenge of identifying churn in a non-contractual context by combining Bisecting K-Means for customer segmentation and Long Short-Term Memory (LSTM) for churn prediction based on monthly revenue. The Bisecting K-Means algorithm produced three clusters with a Davies-Bouldin Index (DBI) of 0.46, indicating effective segmentation. The LSTM model achieved a validation accuracy of 96% and a test accuracy of 95%, with an AUC-ROC of 97%. Results show that cluster 1 has the highest churn rate at 100%, followed by cluster 2 at 16.31%, and cluster 0 with the lowest at 3.21%. Out of 5,163 customers in the test data, 920 were identified as churned.