Machine Learning for Data Science

What you will be able to do

• Frame a business or research question as a supervised or unsupervised learning problem and choose an appropriate model family and success metric.
• Build end-to-end scikit-learn pipelines that bundle preprocessing and the estimator into a single object, so the same steps run identically in training and in production.
• Engineer features from raw tabular data using encoding, scaling, imputation, and ColumnTransformer, and construct interactions and transformations that add real signal.
• Select features deliberately, cutting noise and redundancy with filter, wrapper, and model-based methods instead of feeding the model everything.
• Train and compare linear and regularized models, decision trees, random forests, and gradient boosting, then tune them with cross-validated hyperparameter search.
• Design validation that reflects reality, choosing the right cross-validation scheme and avoiding data leakage and overfitting to the validation set.
• Choose and interpret evaluation metrics that match the decision, including precision, recall, ROC AUC, PR curves, calibration, and regression error, and handle class imbalance honestly.
• Apply clustering and dimensionality reduction (k-means, DBSCAN, PCA) to unlabeled data and judge the results with appropriate diagnostics rather than eyeballing.

What is inside

4 modules, 8 lessons. Each module ends in a graded quiz and most carry an assignment.

1. 01

   The scikit-learn Workflow

   Turn a messy real-world question into a well-posed learning problem, then execute it through scikit-learn's estimator API. This module teaches the fit, predict, and transform contract shared by every model, the difference between supervised and unsupervised problems in code, and the validation discipline that keeps results honest: train and test splits, cross-validation matched to your data, and Pipelines that make data leakage structurally impossible. It is the leak-free foundation every later technique in the course depends on.

   2 lessons · 5 quiz questions

2. 02

   Feature Engineering

   Turn raw tabular data into features a model can actually use. You will wire encoding, scaling, and imputation into a ColumnTransformer and Pipeline, engineer interactions and distribution transforms, then select only the signal, keeping every learned step inside cross-validation so nothing leaks from test into train.

   2 lessons · 5 quiz questions · assignment

3. 03

   Supervised Models and Tuning

   Build the core supervised models and learn when each one earns its place. This module moves from linear and regularized models through decision trees, random forests, and gradient boosting, using the bias and variance tradeoff as your guide, then tunes them honestly with cross-validated search and nested selection that will not quietly overfit your validation data.

   2 lessons · 5 quiz questions

4. 04

   Evaluation and Unsupervised Learning

   A model is not finished until you can prove it works for the decision it will drive. This module teaches you to choose and interpret the metrics that matter for classification and regression, to handle class imbalance and calibration, and to see past a misleadingly high accuracy. You then turn to unsupervised learning, using clustering and dimensionality reduction (k-means, DBSCAN, and PCA) to find structure when there are no labels, and judging those results with honest diagnostics.

   2 lessons · 5 quiz questions · assignment