Customer Churn Analysis and Prediction

Project Overview

This project aims to predict customer churn and segment customers for a telecommunications company. By analyzing customer data, performing clustering, and applying predictive modeling techniques, the team identified high-risk customer groups, key churn drivers, and actionable retention strategies. The project combines unsupervised learning (K-Means clustering) with supervised learning (Artificial Neural Networks) to provide both descriptive and predictive insights.

Table of Contents

• Project Overview
• Data Description
• Methodology
  • Data Preprocessing
  • Clustering Analysis
  • Predictive Modeling
• Key Findings
• Retention Strategies
• Limitations & Future Improvements
• Conclusion
• Directory Structure
• Libraries Used
• Reproducibility

Data Description

The dataset consists of 7,043 customer records with the following features:

• Demographic: Gender, SeniorCitizen, Dependents
• Account Info: Tenure, Contract, MonthlyCharges
• Services: PhoneService, MultipleLines, InternetService
• Target Variable: Churn (Yes/No)

The dataset is complete with no missing values, and the target variable is imbalanced: 1,869 churned vs 5,174 non-churned customers.

Methodology

Data Preprocessing

• Loaded data using Pandas and inspected the structure (head(), info(), shape).
• Converted categorical columns to numeric using Label Encoding.
• Scaled numerical features with StandardScaler for uniformity.
• Split the dataset into training (80%) and testing (20%) sets with stratification to maintain class distribution.
• Applied one-hot encoding to categorical features for machine learning compatibility.

Clustering Analysis

• Normalization: Standardized features to mean 0, standard deviation 1.
• Dimensionality Reduction: Applied PCA to reduce to 2 components for visualization.
• K-Means Clustering:
  • Optimal clusters determined using the Elbow Method → K = 4.
  • Visualized clusters with convex hulls and plotted centroids.
  • Saved cluster labels and centers for further analysis.

Predictive Modeling (ANN)

• Built an Artificial Neural Network (ANN) using Keras:
  • Input layer = number of features
  • Hidden layers = 64 and 32 neurons with ReLU and Dropout (0.3)
  • Output layer = 1 neuron with Sigmoid activation for binary classification
• Trained for 50 epochs, batch size 32, with 20% validation split.
• Evaluated performance using:
  • Accuracy: ~78%
  • ROC AUC: 0.66
  • Confusion Matrix: better performance for non-churners than churners
• Saved trained model as ANN_Churn_Model.h5 for future predictions.

Key Findings

• Customer Segmentation: K-Means clustering revealed 3 meaningful groups:

  1. Loyal and stable customers
  2. Value-sensitive customers
  3. High-risk new customers

• Churn Drivers Identified:

  • Contract type (month-to-month has higher churn)
  • Customer tenure (newer customers are at higher risk)
  • Monthly charges (higher charges increase churn probability)
  • Usage of support services and add-ons

• ANN Insights: Confirmed churn patterns based on service engagement, billing levels, and contract duration.

Retention Strategies

Segment-Specific

• Loyal & Stable Customers: Loyalty rewards, maintain service quality, request feedback.
• Value-Sensitive Customers: Targeted discounts, mid-tier bundle offers.
• High-Risk New Customers: Early engagement campaigns, short-term incentives, encourage long-term contracts.

Model-Driven

• Flag high-risk customers using churn probabilities.
• Tailor retention offers based on key drivers (e.g., billing, service usage).
• Integrate predictions into CRM for proactive interventions.
• Automate retention workflows and monitor metrics continuously.

Limitations & Future Improvements

• Class Imbalance: Fewer churned customers → risk of biased predictions.
• Lack of Qualitative Data: Missing customer satisfaction, complaints, or support interactions.
• Model Interpretability: Complex ANN models may be harder for stakeholders to understand.
• Static Data: Dataset is a snapshot; evolving behaviors may affect model accuracy.

Proposed Solutions:

• Use SMOTE or class weighting to handle imbalance.
• Apply SHAP or feature importance for interpretability.
• Continuously update and retrain models as new data arrives.
• Monitor performance over time and optimize hyperparameters.

Conclusion

This project successfully identified high-risk customers and provided actionable insights into customer churn. Combining clustering with supervised learning allows the company to move from reactive retention to proactive strategies. The findings support informed decision-making, customer loyalty enhancement, and long-term value maximization.

Directory Structure

📁Clustering_Analysis
├── 📁Clustering Analysis Documentation
|   |   ├── Clustering_Analysis.docx
|   |   ├── Clustering_Analysis.pdf
|   ├──📁data
|   |   ├── X_train.csv
|   ├── 📁results
|   |   ├── cluster_center.xlsx
|   |   ├── cluster_label.xlsx
|   |   ├── Cluster_scatter_plot.png
|   |   ├── Clustering_results.xlsx
|   |   ├── Elbow.png
|   ├── clustering_analysis.ipynb
├── 📁Data_Preparation
|   ├── 📁Preprocessed_Data
|   |   ├── preprocessed_data_with_encoding_categorical.csv
|   |   ├── preprocessed_dataset.csv
|   ├── 📁Scaling Techniques Documentation
|   |   ├── Data_Preparation.docx
|   |   ├── Data_Preparation.pdf
|   ├── 📁Testing_Data
|   |   ├── X_test.csv
|   |   ├── y_test.csv
|   ├── 📁Training_Data
|   |   ├── X_train.csv
|   |   ├── y_train.csv
|   ├── data_preparation.ipynb
├── 📁Final_Report
|   ├── Final_Report.docx
|   ├── Final_Report.pdf
├── 📁Predictive_Modeling
|   ├── 📁Predictive Modeling Documentation
|   |   ├── Predictive_Modeling.docx
|   |   ├── Predictive_Modeling.pdf
|   ├── 📁results
|   |   ├── ANN_Churn_Model.h5
|   |   ├── Confusion_Matrix.png
|   |   ├── ROC.png
|   ├── Predictive_Analysis.ipynb
└── README.md

Libraries Used

• Python 3.x
• Pandas
• NumPy
• Scipy (ConvexHull)
• Scikit-learn (LabelEncoder, StandardScaler, train_test_split, KMeans, PCA, classification_report, confusion_matrix, accuracy_score, roc_auc_score, roc_curve)
• tensorflow (Sequential, Dense, Dropout)
• Matplotlib
• seaborn
• OS, Sys

Reproducibility

Follow the steps below to set up the project locally:

# 1. Clone the repository
git clone https://github.com/Arman3747/CHURN-PREDICTION.git
cd CHURN-PREDICTION

# 2. Install dependencies
pip install -r requirements.txt

Project Team:

• Sishir Pandey - Project Manager - LinkedIn
• Fahim Arman - Data Engineer - LinkedIn
• Chen - Data Analyst (Clustering)
• Jitesh Akaveeti - Data Analyst (Predictive Modelling)
• Preeti Khatri - Data Analyst (Predictive Modelling)
• Bishesh Aryal - Business Analyst

---

Thanks For Reading !!!