Telco Customer Churn Prediction

Project Overview

This project analyzes and predicts customer churn for a telecommunications company using the [WA_Fn-UseC_-Telco-Customer-Churn.csv][1] dataset. The goal is to identify customers likely to leave the service, enabling proactive retention strategies.

---

Dataset Summary

• Source: Telco Customer Churn public dataset
• Rows: 7,043 customers
• Target: Churn (Yes/No)
• Features:
  • Demographics: gender, SeniorCitizen, Partner, Dependents
  • Account info: tenure, Contract, PaperlessBilling, PaymentMethod
  • Services: PhoneService, MultipleLines, InternetService, OnlineSecurity, OnlineBackup, DeviceProtection, TechSupport, StreamingTV, StreamingMovies
  • Charges: MonthlyCharges, TotalCharges

---

Workflow

1. Data Preprocessing

• Handled missing values and converted data types.
• Encoded categorical variables (binary and one-hot encoding).
• Engineered features (e.g., tenure groups, total spend).
• Scaled numerical features.
• Addressed class imbalance using SMOTE and class weights.

2. Model Building

• Neural Network:
  • Multi-layer perceptron with dropout regularization.
  • Tuned architecture and hyperparameters.
• Random Forest:
  • Used scikit-learn’s RandomForestClassifier with class weighting.
  • Compared performance with neural network.

3. Evaluation

• Split data into training and test sets (80/20).
• Tracked metrics: Precision, Recall, F1 Score, Accuracy, AUC.
• Sampled and compared model predictions to actual churn outcomes.

4. Results

• Neural Network Final Scores:
  • Precision: ~0.64
  • Recall: ~0.51
  • F1 Score: ~0.57
• Random Forest: Similar or slightly better performance, depending on tuning.
• Interpretation:
  • The model identifies more than half of actual churners with reasonable precision, providing actionable insights for retention teams.

5. Sampling and Validation

• Sampled random customers from the test set.
• Displayed predicted churn, probability, and actual churn status for transparency.