앙상블 모델과 고객 세그먼트 분석 기반 이동통신사 고객 이탈 예측 및 맞춤형 전략

초록

This paper Analyzed a customer churn prediction model and churn factors using data sets from two telecommunications companies, IBM Telco and Cell2Cell. A comparative and evaluative analysis was conducted on a range of machine learning models, including logistic regression, Naive Bayes, K-Nearest Neighbors (KNN), XGBoost, and deep learning models. The ensemble stacking technique emerged as the most effective model, achieving 91.01% accuracy for IBM Telco and 86.43% accuracy for Cell2Cell. An analysis of churn factors revealed that contract terms, usage period, monthly fee, payment method, and availability of additional services were the primary influencing factors. The initial 12 months of service were identified as a pivotal period for mitigating churn. By these findings, a classification system was devised to categorize customers into four distinct segments based on their usage period and fee level. This approach was adopted to formulate customized strategies tailored to each segment. Subsequent analysis revealed the presence of similar churn patterns across both datasets, thereby substantiating the model's generalizability. This study proposes a pragmatic solution that has the potential to assist telecommunications companies in sustaining their profitability and competitiveness. The proposed solution involves the systematic management of customer attrition, a strategy that is informed by data analysis.

키워드