Machine Learning to Identify Bank Customers Interested in a Personal Loan

| Data Science · Projects

Note: This is a project I completed for U.T. Austin’s Post Graduate Program in AI & Machine Learning: Business Applications. The data and business scenarios are part of a simulated case study and do not represent the actual operations of any real-world entity.

Executive Summary

Context

The subject of this analysis is a US-based financial institution with a robust customer base of depositors (liability customers). While the bank maintains a steady flow of deposits, the segment of customers who also hold loan products (asset customers) remains small. To drive revenue through interest income, the bank’s management is focused on converting existing liability customers into personal loan holders while ensuring high retention of their current deposits.

A previous marketing campaign achieved a 9% conversion rate. To build on this success, this project leverages machine learning to identify high-probability prospects, allowing for more efficient, targeted marketing spend and a higher success ratio.

Objective

Data Dictionary

1. Environment Setup & Data Acquisition

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
import warnings

from sklearn.preprocessing import MinMaxScaler
from sklearn.model_selection import train_test_split, GridSearchCV
from sklearn.tree import DecisionTreeClassifier, export_text, plot_tree
from sklearn.metrics import (
    f1_score,
    accuracy_score,
    recall_score,
    precision_score,
    confusion_matrix,
    make_scorer,
)

warnings.filterwarnings("ignore")

# Data ingestion
loan_data = pd.read_csv('Loan_Modelling.csv')
df = loan_data.copy()

2. Data Integrity & Cleaning

The dataset consists of 5,000 observations. We performed standard cleanup, including the removal of non-predictive identifiers and correcting logical inconsistencies in the Experience feature.

# Removing ID as it holds no predictive weight
df = df.drop(["ID"], axis=1)

# Correcting negative 'Experience' values to absolute values
df['Experience'] = df['Experience'].abs()

# Verification of data types and null counts
print(df.info())

3. Exploratory Data Analysis (EDA)

Critical Findings

# Visualizing the relationship between Income, Education, and Loan Acceptance
plt.figure(figsize=(10, 6))
sns.boxplot(x='Education', y='Income', hue='Personal_Loan', data=df)
plt.title('Income Distribution by Education and Loan Acceptance')
plt.show()

4. Feature Engineering & Preprocessing

To capture geographic influence, we mapped ZIP Code prefixes to California’s major economic regions.

def map_regions(zip_code):
    prefix = str(zip_code)[:2]
    mapping = {
        '90': 'Los Angeles', '91': 'Pasadena', '92': 'San Diego',
        '94': 'SF Bay Area', '95': 'San Jose'
    }
    return mapping.get(prefix, 'Other California')

df['Region'] = df['ZIPCode'].apply(map_regions)
df = pd.get_dummies(df, columns=['Region'], drop_first=True)
df.drop(['ZIPCode'], axis=1, inplace=True)

# Train-Test Split (80/20)
X = df.drop('Personal_Loan', axis=1)
y = df['Personal_Loan']
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42, stratify=y)

5. Predictive Modeling: Decision Trees

We utilized a Decision Tree Classifier due to its high interpretability—a critical requirement for banking stakeholders who need to justify marketing logic.

Optimization Strategy

We compared three iterations of the model:

  1. Base Model: Highly complex, exhibiting signs of overfitting.
  2. Pre-Pruned: Optimized via GridSearchCV for maximum recall.
  3. Post-Pruned (Selected): Utilized Cost-Complexity Pruning (ccp_alpha) to create a robust, generalized model.
# Final Model: Post-Pruned Decision Tree
final_model = DecisionTreeClassifier(
    ccp_alpha=0.01, 
    class_weight={0: 0.15, 1: 0.85}, 
    random_state=1
)
final_model.fit(X_train, y_train)

6. Model Evaluation

Metric Training Set Test Set
Accuracy 98.1% 97.5%
Recall 90.1% 89.6%
Precision 90.3% 85.6%

Evaluation Logic: In a loan campaign, Recall is our primary metric. We want to ensure we capture as many potential “Yes” responders as possible, even at the cost of slight precision loss (mailing a few “No” responders).

7. Actionable Insights & Business Strategy

Marketing Targets

Strategic Compliance & Risk

Feature Importance

The visualization below confirms that Income, Education, and Family Size are the primary drivers of the model’s decisions.

# Plotting Feature Importances
importances = final_model.feature_importances_
indices = np.argsort(importances)
plt.figure(figsize=(8, 8))
plt.title("Feature Importances")
plt.barh(range(len(indices)), importances[indices], color="steelblue")
plt.yticks(range(len(indices)), [X_train.columns[i] for i in indices])
plt.show()