I85PREP
⌘ K
IBM Data Science Professional Certificate study materials
IBM Data Science Professional Certificate study resources

IBM Data Science Professional Certificate: Machine Learning with Python

IBM Data Science Professional Certificate · Work through your exam, one topic at a time.

All guide sections

Machine Learning with Python

Machine Learning with Python

This domain tests the scikit-learn workflow: splitting data, building pipelines, training supervised and unsupervised models, and evaluating them with metrics that fit the problem. Expect questions on leakage, imbalanced classes, and overfitting.

Core rules

  • Estimators implement fit, predict, and often transform or fit_transform.
  • Split into train and test before preprocessing; put imputation and scaling inside a pipeline to prevent leakage.
  • Classification metrics include accuracy, precision, recall, F1, ROC-AUC, and the confusion matrix. Regression metrics include MAE, RMSE, and R-squared.
  • Use cross-validation for stable estimates and stratified folds for class imbalance.
  • Regularization and simpler models reduce overfitting; tree ensembles often improve tabular performance.
  • k-means uses distance; evaluate with inertia and silhouette score.

Worked example

A churn model has 5% positives. A constant "no churn" prediction scores 95% accuracy, which is useless. Reasoning: use a stratified train/test split, set class_weight='balanced', and evaluate recall, precision, F1, and the confusion matrix on the holdout set. If missing a churner is expensive, favor recall and tune the decision threshold. Build a pipeline: imputer, scaler, logistic regression, so preprocessing is learned only from training folds.

Traps

  • Scaling or imputing before the train/test split.
  • Tuning hyperparameters on the test set.
  • Reporting accuracy alone with imbalanced classes.
  • Calling fit_transform on test data.
  • Forgetting random_state and getting non-reproducible splits.
  • Leaking target-derived features into training.