Instance-Based Classification Methods

Instance-based classification algorithms perform their main learning process at the instance level. They try to approximate a function that assigns class labels to instances. The instance classifier is combined with an underlying MI assumption, which links the class label of instances inside a bag with the bag class label. Many strategies have been devised to construct the instance classifier. We discuss the most prominent of them: wrapper methods (Sect. 4.2), maximum likelihood methods (Sect. 4.3), decision trees and rules methods (Sect. 4.4), maximum margin methods (Sect. 4.5), connectionist methods (Sect. 4.6), and evolutionary methods (Sect. 4.7). An experimental analysis on the performance of representative instance-based classifiers is presented in Sect. 4.8. Summarizing remarks are given in Sect. 4.9.