Foundations¶
See also
Most of the content in this chapter has been taken from the Ph.D. thesis ‘Multi-label Rule Learning’, Michael Rapp (2022).
Among the most common machine learning approaches, one may distinguish between statistical and symbolic methods. The former use statistical optimization techniques to determine the parameters of a predictive function. Examples include artificial neural networks, support vector machines, or logistic regression. Symbolic learning methods rely on symbolic descriptions to represent learned concepts and capture knowledge about a problem domain. For example, in decision tree learning or rule learning, which this project focuses on primarily, models are typically represented in terms of logical if-then-clauses that test for the properties of given examples to determine a prediction. In the following, we discuss the general methodology of these machine learning methods, as well as the conceptual framework used by our implementations: