Strategi Retensi Pelanggan Berbasis Historis: Optimalisasi Model Prediksi Churn Menggunakan Machine Learning

Abstract

Retensi pelanggan merupakan aspek strategis yang sangat penting dalam menjaga keberlanjutan bisnis, terutama di era kompetisi yang semakin ketat. Penelitian ini berfokus pada upaya optimalisasi model prediksi customer churn berbasis riwayat historis pelanggan dengan memanfaatkan pendekatan machine learning. Dua algoritma utama yang digunakan adalah Support Vector Machine (SVM) dan Jaringan Saraf Tiruan (Artificial Neural Network/ANN) sebagai representasi dari metode ANN. Untuk meningkatkan performa prediksi, diterapkan pula teknik ensemble classifier berupa bagging dan boosting. Guna mengatasi kompleksitas data dan mengurangi risiko overfitting, digunakan teknik dimensionality reduction melalui Principal Component Analysis (PCA). Dataset yang digunakan mencakup berbagai variabel penting seperti data demografis, perilaku pembelian, serta interaksi pelanggan dengan perusahaan. Hasil penelitian menunjukkan bahwa penerapan PCA mampu meningkatkan akurasi model, di mana ANN mencapai 92,37% dan SVM 85,13%. Penerapan metode boosting meningkatkan performa menjadi 93,34% untuk ANN dan 92,73% untuk SVM, sedangkan hasil terbaik diperoleh melalui bagging dengan akurasi 94,38% dan 94,15%. Temuan ini membuktikan bahwa kombinasi antara reduksi dimensi dan ensemble classifier dapat secara signifikan meningkatkan ketepatan prediksi customer churn, sehingga mendukung pengambilan keputusan strategis dan penyusunan strategi retensi pelanggan yang lebih proaktif, terukur, dan efektif. Customer retention is a very important strategic aspect in maintaining business sustainability, especially in an era of increasingly fierce competition. This study focuses on optimizing the customer churn prediction model based on customer historical data by utilizing a machine learning approach. The two main algorithms used are Support Vector Machine (SVM) and Artificial Neural Network (ANN) as representations of ANN methods. To improve prediction performance, ensemble classifier techniques such as bagging and boosting were also applied. To overcome data complexity and reduce the risk of overfitting, dimensionality reduction techniques were used through Principal Component Analysis (PCA). The dataset used included various important variables such as demographic data, purchasing behavior, and customer interactions with the company. The results show that the application of PCA improves model accuracy, with ANN reaching 92.37% and SVM 85.13%. The application of the boosting method improves performance to 93.34% for ANN and 92.73% for SVM, while the best results are obtained through bagging with an accuracy of 94.38% and 94.15%. These findings prove that the combination of dimension reduction and ensemble classifiers can significantly improve the accuracy of churn prediction, thereby supporting strategic decision-making and the development of more proactive, measurable, and effective customer retention strategies.

Keywords

• Retensi
• Strategi
• Machine Learning
• Ensamble Classifer