Investigating the use of tabu search to find near-optimal solutions in multiclassifier systems
Binary trees provide an ideal framework for many decision problems due to their logical, understandable structures and the computational advantages of the “divide and conquer” paradigm. They can be particularly advantageous for classification applications, which involve categorization of information into groups that are in some sense homogeneous. Algorithms used in construction of decision trees used in classification problems are typically greedy. A new algorithm was developed in this study which incorporates Tabu Search (TS) in the feature selection aspect of hierarchical classification trees. Specifically, it is implemented within the hierarchical classification problem framework of the Binary Hierarchical Classifier (BHC) which has been shown to be advantageous for classification problems with a large number of output classes. The algorithm incorporates feature selection as a means for input space and classifier complexity reduction for a static tree; the algorithm was also extended and coupled with the BHC to allow TS feature selection to aid in building the class hierarchy. Finally, a new algorithm was developed which uses TS in the rearrangement of the nodes of a binary classification tree. Since the use of highly accurate classification algorithms is vital in fields such as medical diagnoses, character recognition, target detection, and land cover mapping, the primary goal of this research is to attain improved classification accuracies.