Credit Limit Prediction using Machine Learning

This project focuses on predicting the credit limit of customers using various machine learning models. The dataset consists of customer demographic, financial, and behavioral attributes. The project follows a structured data preprocessing, exploratory data analysis (EDA), model selection, and evaluation approach.

Table of Contents

• Dataset
• Data Preprocessing
• Exploratory Data Analysis (EDA)
• Model Training & Evaluation
• Results
• Dependencies

Dataset

The dataset used in this project is CreditPrediction.csv, which contains financial and demographic details of customers.

Features:

Column	Description
CLIENTNUM	Unique identifier for the customer
Customer_Age	Age of the customer in years
Gender	M (Male) / F (Female)
Dependent_count	Number of dependents the customer has
Education_Level	Customer's education level (e.g., High School, Graduate, Doctorate)
Marital_Status	Marital status of the customer (Married, Single, Divorced, Unknown)
Income_Category	Annual income category (e.g., Less than $40K, $40K-$60K, $60K-$80K, etc.)
Card_Category	Type of credit card (Blue, Silver, Gold, Platinum)
Months_on_book	Number of months the customer has been with the bank
Total_Relationship_Count	Total number of products held by the customer
Months_Inactive_12_mon	Number of months the customer has been inactive in the last 12 months
Contacts_Count_12_mon	Number of contacts made in the last 12 months
Credit_Limit	Credit limit assigned to the customer (Target Variable)
Total_Revolving_Bal	Total revolving balance on the credit card
Total_Amt_Chng_Q4_Q1	Change in transaction amount (Q4 over Q1)
Total_Trans_Amt	Total transaction amount in the last 12 months
Total_Trans_Ct	Total transaction count in the last 12 months
Total_Ct_Chng_Q4_Q1	Change in transaction count (Q4 over Q1)
Avg_Utilization_Ratio	Average card utilization ratio
Unnamed: 19	Unused column removed from the dataset

Data Preprocessing

Handling Missing Values

• Unknown values replaced with NaN.
• Columns like CLIENTNUM and Unnamed: 19 were dropped as they were not useful.
• Missing values were handled using mode imputation for categorical data and mean imputation for numerical data.
• K-Nearest Neighbors (KNN) imputation was also explored.

from sklearn.impute import KNNImputer
imputer = KNNImputer(n_neighbors=3)
df_imputed = pd.DataFrame(imputer.fit_transform(df), columns=df.columns)

Handling Duplicates & Outliers

• Duplicate records were removed.
• Outliers were detected using Interquartile Range (IQR) and either removed or replaced.

Exploratory Data Analysis (EDA)

• Distribution of Card Type against Credit Limits

• Income Category Distribution

• Card Type Distribution

• Education Level Distribution

Model Training & Evaluation

We experimented with multiple regression models:

• Linear Regression
• Ridge Regression
• Random Forest Regressor (Best performing model)
• Gradient Boosting
• K-Means Clustering + Random Forest

from sklearn.ensemble import RandomForestRegressor
model = RandomForestRegressor()
model.fit(X_train, y_train)
y_pred = model.predict(X_test)

Results

Model	Mean Squared Error (MSE)	R-squared (R²)
Linear Regression	33M	0.60
Random Forest	10M	0.87
KMeans + Random Forest	11M	0.84

PyCaret Usage

PyCaret is an open-source, low-code machine learning library that automates model training and evaluation. It provides easy-to-use functions for feature selection, preprocessing, and hyperparameter tuning.

PyCaret in Action

from pycaret.regression import *
s = setup(df_imputed, target="Credit_Limit", session_id=123)
best = compare_models()

Dependencies

• Python 3.8+
• Pandas
• NumPy
• Scikit-learn
• Matplotlib
• Seaborn
• PyCaret