In this, we analyzed loan mispayment dataset from an NBFC company to track customers who are likely to miss the next patment cycle.
Applied data exploration, clustering and predictive modeling to uncover patterns based on which clustered the customers into different persona to drive business growth & create communication strategies for each clusters for CRM.
- Setting Business Objective : Our primary goal is to transition from a generic, manual recovery system to a sophisticated, data-driven behavioral recovery framework.
- Setting Analytical Objectives :
- Predictive Risk Modeling : Classify the probability of next-payment default accurately. Treat missed payments not just as failures, but as crucial early warning signals for proactive intervention.
- Behavioral Segmentation : Classify the probability of next-payment default accurately. Treat missed payments not just as failures, but as crucial early warning signals for proactive intervention.
- Performing Data cleaning
- Feature engineering : Derived features using existing fields
| Features | Formula Applied |
|---|---|
DebtToIncomeRatio |
log(LoanAmount/Income) |
RiskScore |
MissedPayments+(DelaysDays/30)+(Complaints*1.5)d |
MonthlyInstallment |
log((LoanAmount * (1+ ((InterestRate*TenureMonths)/1200)))/TenureMonths)) |
RepaymentBurden |
log(((LoanAmount * (1+ ((InterestRate*TenureMonths)/1200)))/TenureMonths)/(Income/12)) |
SentimentRiskFactor |
log(1+MissedPayments * (1-SentimentScore)) |
InteractionResponsiveness |
log(1+(InteractionAttempts / (ResponseTimeHours+1))) |
DigitalEngagementIndex |
(AppUsageFrequency + (WebsiteVisits/49))/2 |
- Techniques Used :
- Random Forest :
- K-Means Clustering : Dividing into the below 4 Clusters (based on Silhouette analysis & Business use case)
| High Worth Ghost | Vocal Adversary |
| Struggling Scholar | Digital Darling |
