KEY METRICS
98.7%
Recall on holdout data
5,000
Customer records analyzed
60-70%
Potential cost reduction
4
Key predictive features identified
The Challenge
A bank needed to expand its personal loan business by converting existing depositors into borrowers. With a current acceptance rate of only 9.6% (480 of 5,000 customers), broad marketing campaigns were expensive and inefficient. The bank needed a data-driven approach to identify which customers would be most receptive to a personal loan offer.
Approach & Methodology
The project followed a structured ML pipeline: Data Collection → Data Cleaning → Feature Engineering → Exploratory Data Analysis → Model Building → Pruning & Tuning → Evaluation & Selection.
Key steps:
- Analyzed 5,000 customer records across 14 attributes (demographic, financial, banking)
- Applied IQR method for outlier detection
- Used one-hot encoding for categorical variables
- Addressed significant class imbalance (90.4% vs 9.6%)
- Built Decision Tree classifier optimized for recall (minimizing missed opportunities)
Key Findings
The model identified four primary drivers of loan acceptance:
1. Annual Income (64.6% importance)
Acceptors averaged $144,700 vs $66,200 for non-acceptors
2. Household Size (15.8% importance)
Larger families showed higher propensity to accept loan offers
3. Education Level (14.1% importance)
Higher education correlated with increased acceptance rates
4. Average Monthly Credit Card Spend (5.5% importance)
Acceptors spent $3,910/month vs $1,730 for non-acceptors
Interactive Prediction Demo
See How the Model Would Classify Your Customer
Adjust the customer attributes below to see how the model would classify loan acceptance probability based on the key findings above.
$80K
Undergrad
2 members
$2K
Predicted Outcome
-
Confidence Score
-
Business Impact
- Enables highly targeted campaigns reducing costs by 60–70% vs broad outreach
- Dramatically improves conversion rates through precise customer segmentation
- Converts existing depositors into borrowers, expanding loan portfolio
- Identifies high-value segments for focused product development
Tools & Technologies
Python
Decision Trees
Scikit-learn
Feature Engineering
EDA
IQR Outlier Detection
One-Hot Encoding
Classification