Classical machine learning

Status: Classical ML track in progress. Complete the data and mathematics track first.

Prerequisites

• Python variables, conditions, loops, and functions
• pandas tables and basic visualization
• Mean, median, spread, and sampling
• Vectors and dot products

What students will produce

By the end of this stage, students should be able to:

• define a prediction question with features and a target;
• create a simple baseline before choosing a complex model;
• split data without leaking test information;
• select a metric that matches the real question;
• inspect errors instead of reporting only one score; and
• document limitations and possible misuse.

Ordered subtopics

1. Features, labels, training, and inference
2. Baselines and train/test splits
3. Linear regression
4. Classification and logistic regression
5. Confusion matrices and evaluation metrics
6. Overfitting, underfitting, and cross-validation
7. Decision trees and ensembles
8. Nearest neighbors and support-vector machines
9. Clustering and dimensionality reduction
10. Feature engineering and pipelines
11. Error analysis and model cards

Planned project: compare several models on the same dataset and justify the final choice.

Best practices

• Define the prediction moment before selecting features.
• Keep training, validation, and test roles separate.
• Compare every model with a simple baseline.
• Report the metric, units, threshold, sample size, and error costs.
• Inspect examples of failure and document out-of-scope uses.

References

• scikit-learn User Guide
• Google Machine Learning Crash Course

Completion evidence

Submit a notebook containing a dataset description, baseline, train/test split, at least two models, a metric table, an error analysis, and a short model card. A high score is not enough if the test setup is invalid or the result is not appropriate for the problem.