⚡ Naive Bayes for Emergency Triage

In chapter 4 the Minermont team wants a system that suggests the urgency level of each emergency patient, from 1 (red) to 5 (blue), always as a support for the triage nurse. This notebook follows Martha's notes (section 4.5) on the Naive Bayes classifier: it implements it by hand, compares it with scikit-learn and shows where its naive assumption helps and where it hurts.

The interactive notebook includes:

• Bayes' theorem, the prior, the likelihood and the naive independence assumption.
• A prediction worked out by hand from two symptoms.
• Gaussian Naive Bayes by hand, with logarithms, and why a bell curve fails for yes/no symptoms.
• Naive Bayes by counting frequencies, with vital signs cut into intervals and Laplace smoothing, checked against CategoricalNB.
• What happens when the same evidence is counted several times: an overconfident model.
• Final evaluation with the confusion matrix and per-class metrics.

Want to modify the code and experiment on your own?
Open the notebook in Google Colab

Bibliography

• Domingos, P. and Pazzani, M. (1997). On the optimality of the simple Bayesian classifier under zero-one loss. Machine Learning, 29, 103–130. Why Naive Bayes can classify well even when its independence assumption is false. https://doi.org/10.1023/A:1007413511361
• scikit-learn User Guide – Naive Bayes: Gaussian, multinomial, Bernoulli and categorical variants. https://scikit-learn.org/stable/modules/naive_bayes.html
• Manning, C. D., Raghavan, P. and Schütze, H. – Introduction to Information Retrieval, chapter 13: Naive Bayes text classification, logarithms and add-one smoothing. https://nlp.stanford.edu/IR-book/html/htmledition/naive-bayes-text-classification-1.html