Comparison between record to record travel and great deluge attribute reduction algorithms for classification problem

Abstract

In this paper, two single-solution-based meta-heuristic methods for attribute reduction are presented. The first one is based on a record-to-record travel algorithm, while the second is based on a Great Deluge algorithm. These two methods are coded as RRT and m-GD, respectively. Both algorithms are deterministic optimisation algorithms, where their structures are inspired by and resemble the Simulated Annealing algorithm, while they differ in the acceptance of worse solutions. Moreover, they belong to the same family of meta-heuristic algorithms that try to avoid stacking in the local optima by accepting non-improving neighbours. The obtained reducts from both algorithms were passed to ROSETTA and the classification accuracy and the number of generated rules are reported. Computational experiments confirm that RRT m-GD is able to select the most informative attributes which leads to a higher classification accuracy.