Building Robust Credit Scoring Models (Part 3)

2026-03-20 · Source: Towards Data Science · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Intermediate, long

Summary

This article, the third in a series on building robust credit scoring models, details the critical data preprocessing steps of handling outliers and missing values. It uses an open-source Kaggle Credit Scoring Dataset with 32,581 observations and 12 variables to illustrate these techniques. The author first addresses the creation of an artificial "year" variable from `cb_person_cred_hist_length` to enable proper time-based data splitting. The article then explains the importance of splitting data into training (70%), test (30%), and out-of-time (OOT) (2022 data) sets *before* preprocessing to preserve model generalization. Outlier treatment is demonstrated using the IQR method, with a discussion of its applicability and alternatives like Winsorization, especially for variables like `person_age`. Finally, it covers missing value imputation for `person_emp_length` (assigned to 0 for conservative credit scoring) and `loan_int_rate` (imputed with the median), distinguishing between Missing At Random (MAR) and Missing Completely At Random (MCAR) mechanisms.

Key takeaway

For Data Scientists developing credit scoring models, rigorously separating data into training, test, and out-of-time sets *before* any preprocessing is non-negotiable. This ensures that all outlier treatments and missing value imputations are calibrated solely on the training data, preventing data leakage and preserving the model's ability to generalize to new, unseen borrowers. Always replicate these transformations consistently across all datasets to maintain independence and avoid biased performance evaluations.

Key insights

Proper data splitting and preprocessing are crucial for building generalizable and stable credit scoring models.

Principles

• Split data before preprocessing to prevent bias.
• Calibrate preprocessing only on training data.
• Domain expert input is vital for data treatment.

Method

Create a time variable, split data into train/test/OOT, apply IQR-based outlier treatment or Winsorization, and impute missing values based on MAR/MCAR analysis, ensuring all steps are calibrated on training data.

In practice

• Use stratified split for train/test sets.
• Apply IQR method for outlier capping.
• Impute `person_emp_length` with 0 for conservatism.

Topics

• Credit Scoring
• Data Preprocessing
• Outlier Treatment
• Missing Value Imputation
• Model Validation

Best for: Data Scientist, Machine Learning Engineer, AI Student

Related on AIssential

• Exploratory Data Analysis for Credit Scoring with Python — Thorough descriptive data analysis is crucial for understanding credit risk and identifying predictive variables before modeling.
• Data Leakage in Machine Learning: Why You Must Split Before Preprocessing — Preprocessing steps that learn from data must occur after splitting to prevent data leakage and ensure valid model evaluation.
• How to Select Variables Robustly in a Scoring Model — Robust variable selection prioritizes stability across data subsets over performance on a single training set.
• When Linear Regression Beats Random Forest: A UK Salary Prediction Case Study — Model complexity must be earned by data, as simpler models can outperform complex ones on small datasets.
• Your Model Scored 96% Accuracy. Think That’s Success? Think Again. — Model accuracy alone is insufficient; calibrated probabilities and cost-optimized thresholds drive real business value.
• Data Preprocessing in Machine Learning: Working with Numerical & Categorical Data — Data preprocessing is fundamental for machine learning model accuracy, ensuring clean, transformed data for effective learning.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Towards Data Science.