Telecom Customer Churn Analysis Report
- Abhishek ::
- Jun 8
- 5 min read
Updated: Jun 18
Author: Abhishek Sorari
This project analyzes customer behavioral and service usage data from a telecom provider(imaginary) to identify factors influencing churn. The model helps the business proactively identify customers likely to cancel their service, enabling targeted retention strategies.
Table of Contents
Introduction
Customer churn is a critical business challenge in the telecom industry, as retaining customers is typically more cost-effective than acquiring new ones. This analysis explores a Telco Customer Churn dataset to understand which factors drive customers to leave. By leveraging exploratory data analysis (EDA), correlation analysis, and model explainability tools such as SHAP (SHapley Additive exPlanations), we uncover patterns and actionable insights. The following sections present and interpret key visualizations from the analysis, providing a holistic view of churn behavior in this customer base.
Graph 1: SHAP Summary Plot (Beeswarm Plot)
This SHAP summary plot shows the impact of each feature on the model's prediction of churn. Each dot represents a customer. Features like tenure, Contract_Two year, and MonthlyCharges have the largest influence on predictions. Higher tenure and longer contracts reduce the likelihood of churn (blue dots on the negative side), while higher monthly charges increase churn risk (red dots on the positive side). The plot highlights which features the model finds most informative and how different values of these features push predictions higher or lower.
Graph 2: SHAP Feature Importance Bar Plot
This bar plot shows the average magnitude of impact of each feature on churn predictions. Tenure is the most influential feature, followed by contract length and monthly charges. The plot suggests that retention strategies focused on encouraging longer tenure and multi-year contracts may reduce churn. Other factors like payment method and availability of online security also contribute meaningfully. This graph provides a clear ranking of which variables most drive the model's decisions — an actionable guide for targeting business interventions.
Graph 3: SHAP Dependence Plot — MonthlyCharges
This plot shows how individual values of MonthlyCharges affect churn prediction, colored by Contract_Two year. Higher MonthlyCharges tend to increase churn risk, but customers with two-year contracts (red dots) show less sensitivity — they are less likely to churn even at higher charges. This insight implies that bundling higher-cost services with longer contract commitments can mitigate churn risk for high-paying customers. Conversely, customers on month-to-month plans with high charges are at significant churn risk.
Graph 4: SHAP Dependence Plot — Tenure
This SHAP dependence plot shows how tenure impacts churn predictions. Customers with low tenure (< 10 months) strongly increase churn probability (positive SHAP value), while customers with tenure > 50 months significantly lower churn risk. TotalCharges is used as a color gradient — showing that higher TotalCharges generally correlate with longer tenure and lower churn risk. The takeaway: early-stage customers are the highest churn risk, and efforts should focus on onboarding and engagement during the first year.
Graph 5: Correlation Heatmap
The correlation heatmap shows how each feature correlates with churn. Negative correlations (blue) indicate that features like tenure, Contract_Two year, OnlineSecurity_Yes, and TotalCharges are associated with lower churn. Positive correlations (red) suggest that features like InternetService_Fiber optic, PaymentMethod_Electronic check, and MonthlyCharges are linked to higher churn. This matrix gives a broad overview of linear relationships in the data, useful for both intuition and feature selection. It confirms patterns seen in the SHAP analysis.
Graph 6: Monthly Charges by Churn Status
This density plot compares MonthlyCharges distribution across churned and non-churned customers. Churned customers (orange) tend to cluster more heavily at higher monthly charges (> $70), while non-churned customers (blue) are spread across lower and mid-range charges. This confirms that high billing amounts contribute to churn risk — potentially due to affordability issues or dissatisfaction with perceived value. Managing pricing and customer expectations around high-value services could reduce churn in this segment.
Graph 7: Tenure Distribution by Churn
This histogram shows the count of customers by tenure and churn status. Churn is highest among customers with very short tenure (0–10 months), consistent with onboarding challenges and early dissatisfaction. Conversely, long-tenure customers (60–72 months) almost never churn — showing strong customer loyalty once established. The sharp contrast between these groups underlines the importance of retention efforts during the first year of customer lifecycle. Improving onboarding experience and addressing early pain points can have a major impact on reducing churn.
Graph 8: Tenure vs Monthly Charges by Churn Status
This scatterplot visualizes MonthlyCharges vs Tenure, colored by churn. The top-left region (low tenure, high charges) is dominated by churned customers, confirming previous insights. In contrast, long-tenure customers (right side of the plot), even with high MonthlyCharges, rarely churn. This again emphasizes that length of relationship strongly moderates churn risk, and high-spending customers can be retained effectively if relationship longevity is built.
Graph 9 : SHAP Force Plot (Individual Prediction)
This SHAP force plot explains a single customer’s churn prediction. Key churn drivers here are MultipleLines_Yes, StreamingTV_Yes, and a moderate TotalCharges, pushing the prediction toward churn. However, strong retention factors — tenure = 72, Contract_Two year, OnlineSecurity_Yes, and OnlineBackup_Yes — pull the prediction back toward no churn. The net effect is a low predicted churn probability. This plot is a great example of how individual factors combine to produce a final model decision.
Conclusion
Graph No. | Graph Title | One-Liner Conclusion |
1 | SHAP Summary Plot (Beeswarm Plot) | Tenure, contract length, and charges are key churn drivers. |
2 | SHAP Feature Importance Bar Plot | Tenure is the most influential feature on churn. |
3 | SHAP Dependence Plot — MonthlyCharges | Long contracts reduce churn risk even with high charges. |
4 | SHAP Dependence Plot — Tenure | Short-tenure customers are at highest churn risk. |
5 | Correlation Heatmap | Longer tenure, security services reduce churn; high charges increase it. |
6 | Monthly Charges by Churn Status | High monthly charges correlate with higher churn rates. |
7 | Tenure Distribution by Churn | Churn peaks in early tenure (<10 months), drops sharply after. |
8 | Tenure vs Monthly Charges by Churn Status | Short-tenure, high-charge customers are most churn-prone. |
9 | SHAP Force Plot (Individual Prediction) | A customer’s churn is driven by both service and contract factors. |
The visual analysis clearly shows that tenure is the most critical factor influencing churn: customers who stay longer are much less likely to leave. Short-tenure customers with high monthly charges are at the greatest churn risk and should be targeted with proactive retention efforts. Contract type also matters — long-term contracts reduce churn significantly, even among high-paying customers. Security services and add-ons further help retain customers. SHAP analyses offer transparency and validate these patterns. In sum, strategies focused on improving early customer experience, offering attractive long-term contracts, and managing pricing perception will be the most effective in reducing churn.
Comments