Build a leak-free sklearn pipeline
Overview
This question evaluates a candidate's ability to construct a leak-free scikit-learn pipeline encompassing column-wise preprocessing, imbalanced binary classification handling, stratified cross-validation, hyperparameter tuning, and model persistence.
Boston Consulting Group
Oct 13, 2025, 9:49 PM
Data Scientist
Take-home Project
Machine Learning
1
0
Loading...
Take-home: Imbalanced Binary Classification Pipeline with scikit-learn
You are training a binary classifier on tabular data with the following feature schema:
- Numeric: age, balance, n_logins_30d, minutes_watched_7d
- Categorical: country, device_type, plan
Positives are rare (~5%). Complete the tasks below using scikit-learn.
Assume you have a pandas DataFrame df with those feature columns and a binary target column target (1 = positive, 0 = negative).
Tasks
- Split df into train/validation with stratification on target.
-
Build a scikit-learn Pipeline with a ColumnTransformer:
- Numeric → SimpleImputer(strategy='median') then StandardScaler
- Categorical → OneHotEncoder(handle_unknown='ignore', min_frequency=0.01)
- Handle class imbalance via class_weight='balanced'. Briefly explain when you would instead use resampling (e.g., under/over-sampling).
-
Tune at least two models using StratifiedKFold and GridSearchCV with scoring='average_precision':
- LogisticRegression
- GradientBoosting-type model (e.g., HistGradientBoostingClassifier)
- Ensure all preprocessing resides inside the Pipeline so no data leakage can occur during cross-validation.
- Report the best model and its validation metrics: precision, recall, and average precision (AP). Show how to persist the trained pipeline for inference.
Include code snippets and short justifications for key choices.
Solution
Login required
Show