Supervised Instance-based Learning Using Patterns in a Trie like Structure

A persistent trie1-like tree structure is presented as the basis to implement a new classification algorithm of the instance-based learning type that we call Trie-CLASS. Records from a training file are converted into pattern vectors associated with a known class label, forming a hypothesis data model stored permanently into the trie. As a result, disjoint subsets as well as areas of conditional entropy in terms of the class are formed. Classifying new examples of an unknown class is done first, converting the input vector into a pattern; secondly, extracting two best hypotheses patterns from the tree without having to examine a larger subset of possible solutions. Thirdly, compare these patterns with the target and selecting the best option. Classification tests done on several data files have shown accurate results.