Granular Computing in Decision Approximation

"Granular Computing in Decision Approximation" is a specialized book that deals with knowledge discovery in data, where knowledge is understood as a set of relationships between objects and their properties. The focus is on the induction of implicative decision rules within a paradigm based on the processing of granules defined by rough inclusions. These concepts are examined within the framework of rough mereology, an extended version of classical mereology. The book defines fundamental classes of rough inclusions and highlights methods for inducing granular structures from data. The resulting structures are subjected to various classification algorithms, such as k-nearest neighbors and Bayesian classifiers. The detailed experimental results, presented in both tabular and visual form, are derived from fourteen datasets from the UCI data repository. A notable feature of the granular classifiers is that they preserve the accuracy of the original data while significantly reducing both the size of the granulating dataset and the number of granular decision rules. This makes the approach particularly attractive when a small number of rules with high classification accuracy is desired. The fundamental algorithms are explained in the text and illustrated with practical examples, making the book also suitable as a textbook.