Project: Big Data Analysis & Survival Analysis

Student: 马硕峰 | ID: 12313113

Part 2: Survival Analysis (核心展示)

1. Introduction

This project applies survival analysis to a telecom customer churn dataset. It uses Kaplan-Meier estimation, Cox Proportional Hazards model, and AFT model to identify key churn factors and build a Customer Lifetime Value (CLV) evaluation tool.

2. Data & Cohort

Dataset: IBM telecom churn data (7,043 customers)
Cohort: 3,351 customers with month-to-month contracts and internet services.

3. Kaplan-Meier Survival Analysis

• Median customer survival time: ~34 months
• OnlineSecurity, TechSupport, and InternetService show strong separation in survival curves
• Log-rank tests confirm significant differences between groups

4. Cox Proportional Hazards Model

• Key predictors: Dependents, InternetService, OnlineBackup, TechSupport
• DSL customers have ~80% churn risk of Fiber customers
• Proportional hazards assumption partially violated for some variables

5. Accelerated Failure Time (AFT) Model

• Distribution: Log-Logistic
• DeviceProtection, OnlineSecurity, and TechSupport significantly extend customer tenure

6. CLV Dashboard & Business Insights

• Interactive dashboard for CLV calculation
• Supports CAC budgeting and customer segmentation decisions
• Helps identify high-risk churn groups for retention strategies

Website deployed via GitHub Pages | Domain: mashuofeng.me | HTTPS enabled