Katkat
Back

Deep-dive XGBoost handling and overfitting

Capital OneData Engineer · Technical Screen
MediumData EngineerTechnical ScreenMachine Learning

Last updated

This question evaluates proficiency with gradient-boosted decision trees and related competencies such as native versus imputation handling of missing values, causes and control of overfitting via regularization and hyperparameters, selection of metrics and validation strategies for imbalanced outcomes, and practical debugging concerns like data leakage, time-based splits, and calibration for a Data Engineer role. It is commonly asked in Machine Learning interviews to assess both conceptual understanding of algorithm behavior and practical application of model evaluation and deployment-ready validation techniques.

Technical / ML Deep Dive

You used gradient-boosted decision trees (e.g., XGBoost/LightGBM) for a credit risk or response prediction problem.

Answer the following:

  1. Missing values: How do boosted trees handle missing values during training/inference? What options do you have (native handling vs imputation), and when would you choose each?
  2. Overfitting control: What are the main causes of overfitting in boosted trees, and what techniques/hyperparameters would you use to reduce it?
  3. Evaluation: Which metrics would you use for an imbalanced credit outcome (e.g., default), and how would you validate the model to ensure it generalizes?

Be prepared to discuss practical pitfalls (data leakage, time-based splits, calibration) and how you would debug issues.