Hyperboxes and Ant Colony Optimization Applied to Clustering and Classification

The possibilities of using hyperboxes and the ant colony optimization (ACO) metaheuristic to clustering and classification are examined. A clustering method, called HACO (hyperbox clustering with ant colony optimization), is presented for classifying unlabeled data using hyperboxes and the ACO meta-heuristic. It acknowledges the topological information (inherently associated to classification) of the data while looking in a small search space, providing results with high precision in a short time. It is validated using artificial 2D data sets and then applied to a real medical data set, automatically extracting medical risk profiles, a laborious operation for doctors. Clustering results show an improvement of 36% in accuracy and 7 times faster processing time when compared to the usual ant colony optimization approach. It can be further extended to hyperbox shape optimization (fine tune accuracy), automatic parameter setting (improve usability), and applied to diagnosis decision support systems. A method called HACO2 (hyperbox classifier with ant colony optimization type 2) is presented, as a natural extension of HACO, aiming to evolve a hyperbox classifier. It reshapes one or more hyperboxes in a near-optimal way, improving the accuracy of the classifier while maintaining its characteristics of acknowledging the topological information of the data and straightforward interpretation. HACO2 also allows a feature discriminating ratio to indicate which characteristics should be considered as more important for distinguishing classes. It is validated using a artificial 2D data set which resulted in over 90% accuracy. HACO2 is also applied to the benchmark iris data set (4 features), providing results with over 93% accuracy, and to the MIS data set (11 features), with almost 85% accuracy. The two most discriminative features obtained from the method are used in simplified classifiers which result in accuracies of 100% for the iris and 83% for the MIS data sets. Further modifications (automatic parameter setting), extensions (initialization short comes) and applications to customer profiling in the field of telecommunications are discussed. Finally, the HACO3 (hyperbox clustering and classification with ant colony optimization type 3) method is proposed for clustering unlabeled data with hyperboxes and evolving the results to provide a clustering and classification solution. It is a hybrid of the methods HACO and HACO2, which combines their complementary characteristics. It is applied to an artificial data set for analysis and to the iris data set for validation, providing 73.3% accuracy for a full classifier and 95.3% for a simplified version. Possible extensions and implementations of HACO3 are examined