Machine learning
This subject builds the practical judgment needed to turn business tabular data into useful supervised-learning models. It covers the full path from framing a prediction problem to evaluating whether a model is safe and valuable enough to use.
- created
- 28 Aug 2026
- status
- active
- lessons
- 7
- references
- 2
Success looks like
- Translate a business question into examples, features, a target, and a prediction time
- Establish a simple baseline before training a model
- Split data without leakage and explain why untouched test data matters
- Build a reproducible scikit-learn pipeline for numerical and categorical columns
- Choose and interpret classification or regression metrics in business terms
- Compare models with cross-validation and diagnose underfitting or overfitting
- Produce a compact model card with limitations, monitoring needs, and a go or no-go recommendation
Constraints
- Assume basic Python but no prior machine-learning knowledge
- Use business tabular data and scikit-learn
- Keep lessons short, interactive, and centered on decisions rather than algorithm derivations
Out of scope
- Deep learning, computer vision, and language models
- Mathematical proofs of optimization algorithms
- Production infrastructure beyond evaluation, handoff, and monitoring principles
Journey
- 1Frame the predictionTurn a business request into rows, features, target, prediction time, and baselineopen
- 2Split before learningDesign train, validation, and test partitions without leakageopen
- 3Fit the first modelTrain and read a minimal scikit-learn classifieropen
- 4Choose the metricMatch classification metrics and thresholds to business error costsopen
- 5Build the pipelinePreprocess mixed columns inside one reproducible pipelineopen
- 6Compare honestlyUse cross-validation and diagnose fit from train-validation gapsopen
- 7Make the go decisionWrite a model card and decide whether evidence supports useopen