SEGMENTASI DAN PREDIKSI LOYALITAS PELANGGAN BERBASIS RFM DAN NET PROMOTER SCORE (NPS) PROXY PADA PT. XYZ DI MOJOKERTO
CUSTOMER SEGMENTATION AND LOYALTY PREDICTION BASED ON RFM AND NET PROMOTER SCORE (NPS) PROXY AT PT. XYZ IN MOJOKERTO
Nama : Alina Rusyda Hariadi
NIM : 22051214003
Jurusan : Sistem Informasi
Fakultas : Teknik
Nama Lembaga : Universitas Negeri Surabaya
Pembimbing : Dr. Wiyli Yustanti, S.Si., M.Kom.
Loyalitas pelanggan merupakan faktor penting bagi perusahaan dalam mempertahankan keberlangsungan bisnis dan meningkatkan daya saing. PT. XYZ sebagai perusahaan distributor Air Minum Dalam Kemasan (AMDK) perlu memahami perilaku pelanggan untuk menentukan strategi pengelolaan pelanggan yang tepat. Penelitian ini bertujuan untuk melakukan segmentasi dan prediksi loyalitas pelanggan berdasarkan perilaku transaksi menggunakan pendekatan Recency, Frequency, dan Monetary (RFM) serta Net Promoter Score (NPS) Proxy.
Data yang digunakan merupakan data transaksi pelanggan PT. XYZ periode April 2024 hingga April 2025. Pada tahap awal, dilakukan proses data preparation untuk membentuk nilai RFM sebagai representasi perilaku pelanggan. Selanjutnya, segmentasi pelanggan dilakukan menggunakan metode clustering, yaitu K-Means, Agglomerative Hierarchical Clustering, dan DBSCAN, yang kemudian dievaluasi menggunakan Silhouette Score, Calinski-Harabasz Index, dan Davies-Bouldin Index. Hasil evaluasi menunjukkan bahwa metode Agglomerative Hierarchical Clustering memberikan kualitas cluster terbaik.
Hasil segmentasi kemudian dipetakan ke dalam kategori loyalitas pelanggan menggunakan pendekatan NPS Proxy, yaitu Detractor, Passive, dan Promoter. Berdasarkan pemetaan tersebut, dibangun model klasifikasi untuk memprediksi loyalitas pelanggan menggunakan beberapa algoritma, di antaranya KNN, SVC, Gradient Boosting, Logistic Regression, dan Random Forest. Hasil pengujian menunjukkan bahwa Random Forest memberikan performa terbaik pada data uji berdasarkan nilai F1-macro serta keseimbangan antara precision dan recall.
Sebagai hasil akhir, model clustering dan klasifikasi diimplementasikan ke dalam aplikasi berbasis web CUSTIFY menggunakan framework Streamlit. Aplikasi ini mampu menampilkan analitik pelanggan, visualisasi data, serta memprediksi kategori loyalitas pelanggan secara otomatis, sehingga dapat digunakan sebagai alat pendukung pengambilan keputusan oleh perusahaan.
Kata kunci: RFM, Loyalitas Pelanggan, NPS Proxy, Clustering, Random Forest.
Author : Alina Rusyda Hariadi
NIM : 22051214002
Study Program : Bachelor’s Degree of Information Systems
Faculty :Faculty of Engineering
Institution : Surabaya State University
Preceptor : Dr. Wiyli Yustanti, S. Si., M.Kom.
Customer loyalty is an important factor for companies in maintaining business sustainability and improving competitiveness. PT. XYZ, as a distributor of bottled drinking water (AMDK), needs to understand customer transaction behavior in order to determine appropriate customer management strategies. This study aims to perform customer segmentation and predict customer loyalty based on transaction behavior using the Recency, Frequency, and Monetary (RFM) approach and the Net Promoter Score (NPS) Proxy.
The data used in this study consist of customer transaction records of PT. XYZ from April 2024 to April 2025. In the initial stage, data preparation was conducted to generate RFM values as representations of customer behavior. Customer segmentation was then performed using several clustering methods, namely K-Means, Agglomerative Hierarchical Clustering, and DBSCAN. The clustering results were evaluated using the Silhouette Score, Calinski–Harabasz Index, and Davies–Bouldin Index. The evaluation results indicate that Agglomerative Hierarchical Clustering produces the best cluster quality.
The resulting customer segments were subsequently mapped into customer loyalty categories using the NPS Proxy approach, consisting of Detractor, Passive, and Promoter. Based on this mapping, a classification model was developed to predict customer loyalty using several algorithms, including K-Nearest Neighbor (KNN), Support Vector Classifier (SVC), Gradient Boosting, Logistic Regression, and Random Forest. The experimental results show that the Random Forest model achieves the best performance on the test data, as indicated by the highest F1-macro score and a balanced precision and recall across classes.
As a final outcome, the selected clustering and classification models were implemented into a web-based application named CUSTIFY using the Streamlit framework. The application provides customer analytics, data visualization, and automatic customer loyalty prediction, thereby serving as a decision support tool for the company.
Keywords: RFM, Customer Loyalty, NPS Proxy, Clustering, Random Forest