Using Ant Colony Optimization to Build Cluster-Based Classification Systems

Learning cluster-based classification systems is the process of partitioning a training set into data subsets (clusters), and then building a local classifier for each data cluster. The class of a new instance is predicted by first assigning the instance to its nearest cluster, and then using that cluster's local classification model to predict the instance's class. In this paper, we use the Ant Colony Optimization (ACO) meta-heuristic to optimize the data clusters based on a given classification algorithm in an integrated cluster-with-learn manner. The proposed ACO algorithms use two different clustering solution representation approaches: instance-based and me doi d-based, where in the latter the number of clusters is optimized as part of the ACO algorithm's execution. In our experiments, we employ three widely-used classification algorithms, k-nearest neighbours, Ripper, and C4.5, and evaluate performance on 30 UCI benchmark datasets. We compare the ACO results to the traditional c-means clustering algorithm, where the data clusters are built prior to learning the local classifiers.