Title :
Ant Colony Prototype Reduction Algorithm for kNN Classification
Author :
Miloud-Aouidate, A. ; Baba-Ali, A.R.
Author_Institution :
Lab. of Robot., Univ. of Sci. & Technol. Houari Boumediene, Algiers, Algeria
Abstract :
Ant colony optimization (ACO) techniques were proposed to solve many combinatorial problems. As prototype selection (PS) is a combinatorial problem we attempt, in this paper, to address it using an Ant Colony algorithm. This work proposes an Ant Prototype Reducing algorithm (Ant-PR). The goal of this algorithm is to reduce the training set of the 1NN classifier. We compared the Ant-PR performances to two classical well-known kNN condensing algorithms. The results provide evidence that: (1) Ant-PR is competitive with the well-known kNN algorithms, (2) The condensed sets computed by Ant-PR offers better classification accuracy then those obtained by the compared algorithms.
Keywords :
ant colony optimisation; combinatorial mathematics; data reduction; pattern classification; pattern clustering; 1NN classifier; ACO techniques; ant colony prototype reduction algorithm; ant-PR performances; combinatorial problem; condensed sets; k-nearest neighbor classification rule; kNN classification; prototype selection; training set; Accuracy; Algorithm design and analysis; Benchmark testing; Classification algorithms; Genetic algorithms; Prototypes; Training; Ant Colonies Optimisation; Condensing; KNN; Prototype Reduction; Prototype Selection;
Conference_Titel :
Computational Science and Engineering (CSE), 2012 IEEE 15th International Conference on
Conference_Location :
Nicosia
Print_ISBN :
978-1-4673-5165-2
Electronic_ISBN :
978-0-7695-4914-9
DOI :
10.1109/ICCSE.2012.47
