Prediksi Churn Pelanggan di Layanan Streaming Berbasis Analisis Perilaku dan Sentimen dengan Ensemble Machine Learning: Studi Platform Lokal Vidio vs. Netflix Indonesia

Abstract

The streaming service industry in Indonesia faces high challenges in retaining customers (churn rate 15-20%). This study develops a predictive churn model by combining customer behavior and sentiment analysis. this study purpose to compare the dominant churn factors in local (Vidio) and global (Netflix) platforms and build an ensemble machine learning model for accurate prediction. The methods is Analysis of a dataset of 500,000 users (2022-2023) using stacking ensemble techniques (XGBoost + LSTM) and IndoBERT text processing. The model achieved 89% accuracy (F1-score) with the main churn factors: payment issues (Vidio) and content relevance (Netflix). Model-based intervention reduced the churn rate by 25%. Integration of behavioral and sentiment data significantly improves churn prediction performance in the unique Indonesian market.