Customer Churn Prediction Analysis

CIS 9660 - Group 2 Project Proposal

2026-03-16

Team Members

Our Team

  • Harrison Cabe
  • Raúl J. Solá Navarro
  • Samuel Spitzer
  • Victor Murra Schott

Course

CIS 9660 - Data Mining

Proposal Due: March 20, 2026
Presentation: March 16, 2026

Project Overview

The Big Question

Can we predict whether a telecom customer will churn based on their contract type, monthly charges, and service usage?


Dimension Focus
📊 Domain E-commerce & Retail - Telecom Customer Retention
🎯 Task Type Binary Classification (Churn: Yes / No)
🧪 Primary Method Logistic Regression with Interaction Terms

Dataset

Kaggle: Telco Customer Churn

🔗 kaggle.com/datasets/blastchar/telco-customer-churn

Property Value
Observations 7,043 customers
Variables 21 features
Target Churn (Yes / No)

Feature Groups

Demographics
Gender, Senior Citizen, Partner, Dependents

Services
Phone, Internet, Streaming, Security

Account
Contract, Tenure, Monthly/Total Charges, Payment Method

Key Variables

🎯 Target

Churn - Binary: Yes / No

📌 Primary Predictors

  • Contract - Month-to-month, One year, Two year
  • MonthlyCharges - Continuous (USD)
  • InternetService - DSL, Fiber Optic, None
  • tenure - Months with company

📎 Additional Features

  • TotalCharges
  • PaymentMethod
  • PaperlessBilling
  • OnlineSecurity, TechSupport
  • StreamingTV, StreamingMovies

Research Question


Can we predict customer churn from
contract type and monthly charges?


🔮 Predictive

Build a model that accurately classifies customers as likely to churn or stay

🔬 Inferential

Does the interaction between InternetService × tenure significantly affect churn probability?

Methods: EDA

Step 1 - Exploratory Data Analysis

📦 Boxplots
Compare MonthlyCharges and tenure distributions across churn classes

🔵 Scatterplots
Explore relationships between continuous variables (MonthlyCharges, tenure, TotalCharges) and churn

📊 Bar Charts
Churn rates segmented by ContractType, InternetService, and PaymentMethod


Tools: pandas, matplotlib, seaborn

Methods: Logistic Regression

Step 2 - Binary Logistic Regression

  1. Baseline model - ContractType + MonthlyCharges + InternetService + tenure

  2. Interaction term - ContractType × InternetService to test synergy effects on churn probability

  3. Model evaluation - Confusion matrix, Accuracy, Precision, Recall, F1-Score, ROC-AUC

  4. Cross-validation - k-fold CV to assess generalizability


Tools: scikit-learn, statsmodels

Expected Findings

📈 Hypotheses

  • Month-to-month contract customers will have significantly higher churn rates
  • Higher MonthlyCharges correlates with increased churn risk
  • Fiber optic + short tenure customers may show elevated churn (interaction effect)

📐 Model Performance Targets

Metric Target
Accuracy ≥ 80%
ROC-AUC ≥ 0.80
Recall (Churn=Yes) ≥ 75%

Recall prioritized - cost of missing a churner is high

Practical Implications


🎯 Targeted Retention

Send personalized offers - e.g., upgrade a month-to-month user to a discounted annual contract

💰 Pricing Strategy

Restructure pricing tiers or offer loyalty discounts to high-charge, high-risk customers

📈 CLV Optimization

Retaining customers longer directly increases Customer Lifetime Value

🔧 Service Improvement

High churn among fiber optic users may signal service quality issues to investigate

Project Timeline

Milestone Date Deliverable
🗣️ In-Class Presentation March 16, 2026 Slides (email by March 15)
📋 Project Proposal March 20, 2026 1–2 page writeup (Brightspace)
💻 Final Presentation May 11, 2026 10-min talk + Q&A
📄 Final Report May 15, 2026 4–5 page report + Python code + data
⭐ Peer Evaluation May 15, 2026 0–5 scale per teammate

Summary


Dataset: Telco Customer Churn (Kaggle)
7,043 obs × 21 variables

Question: Can contract type & monthly charges predict churn?

Method: Logistic Regression + Interaction Terms
ContractType × InternetService

Goal: Help marketing teams proactively retain at-risk customers