February 11, 2026 https://classroom.github.com/a/jj5OrKRX

References

Crash Course in Classification

XKCD classification

The building blocks of ML are algorithms for regression and classification:

Regression: predicting continuous quantities
Classification: predicting discrete class labels (categories)

Classification Methods

Classification algorithms learn to assign labels to data points based on their features. In health data, this might mean predicting whether a patient has a disease based on lab results.

Reference Card: Common Classification Methods

Method	Description	Strengths
Logistic Regression	Linear model for binary outcomes	Easy to interpret, fast
Decision Trees	Tree structure, splits data by feature values	Intuitive but can overfit
Random Forest	Many decision trees combined	More robust, less overfitting
Support Vector Machines (SVM)	Finds the best boundary between classes	Good for complex data
Naive Bayes	Probabilistic, assumes features are independent	Fast and simple
Neural Networks	Layers of nodes, can model complex patterns	Powerful but less interpretable

it’s all statistics

Model Evaluation

There are many more classification approaches than data scientists, so choosing the best one for your application can be daunting. Thankfully, all of them output predicted classes for each data point. We can use this similarity to define objective performance criteria based on how often the predicted class matches the underlying truth.

I get in trouble with the data science police if I don’t include something about confusion matrices:

Evaluation metrics

Precision (Positive Predictive Value) = $\frac{TP}{TP + FP}$

How well it performs when it predicts positive
Recall (Sensitivity, True Positive Rate) = $\frac{TP}{TP+FN}$

How well it performs among actual positives