Naïve Bayes classifier
Paweł Cichosz · 2015
The naïve Bayes classifier is one of the simplest approaches to the classification task that is still capable of providing reasonable accuracy. Bayesian inference, of which the naïve Bayes classifier is a particularly simple example, is based on the Bayes rule that relates conditional and marginal probabilities. There are two major approaches in applying Bayesian inference to the classification task: model-probability inference and class-probability inference. The naïve Bayes algorithm may need some minor enhancements before it is ready to work using real-world data; the chapter reviews the most important practical issues that need to be taken care of. Missing attribute values are likely to decrease model quality for any modeling algorithm, when occurring for training instances, or classification accuracy, when occurring for classified instances. The not-so-naïve versions of the naïve Bayes classifier substantially increase the computational complexity of model creation and prediction.