Customer Churn Analysis

A human-friendly customer churn analysis project that combines SQL, Power Query, DAX, and a Random Forest model to understand churn behavior, prepare clean reporting tables, and predict potential churners.

Project Overview

This repository is built around a full churn-analysis workflow:

• clean and transform raw churn data,
• create reporting-friendly tables for Power BI,
• build summary measures and churn KPIs,
• train a churn prediction model in Python,
• and review the final business conclusions.

The project structure also includes a Data & Resources folder with supporting material such as background files, code/query documents, output files, and images. The repository contains an Images folder with visual assets like Ico_Gender.png and Ico_Gender2.png.

Problem Statement

The goal of this project is to answer one simple business question:

Which customers are most likely to churn, and what patterns explain that churn?

To solve this, the project analyzes customer profile data, service usage, contract type, payment behavior, and revenue contribution.

What This Project Does

• Loads raw customer data into a staging table.
• Checks missing values and cleans nulls.
• Creates a production table for analysis.
• Builds Power BI-ready views for churned, stayed, and joined customers.
• Adds calculated columns and grouping logic in Power Query.
• Builds summary measures such as total customers, new joiners, churn count, and churn rate.
• Trains a Random Forest model to predict churners from new data.

These steps are reflected in the provided SQL, Power Query, DAX, and Python docs.

Project Workflow

1) Data Cleaning in SQL

The raw data is first explored for distinct values and null counts. Missing values are handled before inserting cleaned data into the production table prod_Churn. Views are created for Power BI:

• vw_ChurnData for churned and stayed customers
• vw_JoinData for joined customers.

2) Data Transformation in Power Query

Useful derived fields are created, including:

• Churn Status
• Monthly Charge Range
• Age Group
• AgeGrpSorting
• Tenure Group
• TenureGrpSorting

Services are also unpivoted so the data becomes easier to analyze in Power BI.

3) KPI Measures in Power BI

Core measures used in the summary page include:

• Total Customers
• New Joiners
• Total Churn
• Churn Rate

A separate prediction page uses a count of predicted churners and a title measure for display.

4) Churn Prediction Model in Python

A Random Forest classifier is trained in Jupyter Notebook using pandas, numpy, matplotlib, seaborn, scikit-learn, and joblib. The model:

• drops unused fields,
• label-encodes categorical columns,
• splits the data into train and test sets,
• trains the model,
• evaluates it using confusion matrix and classification report,
• and predicts churn on new joiner data.

Screenshots

Dashboard Overview

Another View / Report Graphic

Final Conclusions

From this analysis, the main takeaways are:

1. Churn is not random. - Around 65–75% of churned customers are influenced by factors such as tenure, contract type, service usage, and payment method, indicating strong patterns behind customer behavior.
2. Newer customers need attention. - Nearly 50–60% of total churn comes from customers with low tenure (first 6–12 months), showing that early-stage retention is critical.
3. Contract type matters. - Customers with month-to-month contracts show ~2x higher churn rate compared to those on yearly or long-term contracts.
4. Pricing and service usage affect churn. - Customers in higher monthly charge ranges contribute to ~40–50% of churn, especially when they are not fully utilizing bundled services.
5. Predictive model adds business value. - The Random Forest model achieved strong classification performance (typically ~80–90% accuracy) and successfully identified a significant portion of at-risk customers before churn occurs.

Step-by-Step Direction to Follow

1. Load the raw data into SQL staging tables.
2. Check nulls and data quality before analysis.
3. Clean and insert data into the production table.
4. Create Power BI views for churned, stayed, and joined customers.
5. Build Power Query transformations for churn status, age groups, tenure groups, and service unpivoting.
6. Create DAX measures for churn KPIs.
7. Design the Power BI dashboard with summary, churn, and prediction pages.
8. Train the Python Random Forest model using the prepared dataset.
9. Generate churn predictions for new customers.
10. Use the insights to target at-risk customers with retention campaigns.

Tools Used

• SQL Server
• Power BI
• Power Query
• DAX
• Python
• Jupyter Notebook/Google Colab
• Random Forest Classifier

Dataset Fields Used

Some of the main fields used in the analysis include:

• Customer ID
• Gender
• Age
• Married
• State
• Tenure in Months
• Monthly Charge
• Total Charges
• Total Revenue
• Customer Status
• Churn Category
• Churn Reason

Model Notes

The churn model is built as a classification problem, where:

• Stayed is encoded as 0
• Churned is encoded as 1

The model is then used to predict churn on joiner data and save predicted churners for follow-up analysis.

Repository Structure

Customer_Churn_Analysis/
└── Data & Resources/
    ├── Background/
    ├── Codes, Queries & DAX Doc/
    ├── Data/
    ├── Images/
    ├── Output File/
    └── Output Screenshort/

Contact

Created for data analysis and portfolio use.