Authors:
Khalid M. Salama
1
and
Ashraf M. Abdelbar
2
Affiliations:
1
University of Kent, United Kingdom
;
2
Brandon University, Canada
Keyword(s):
Ant Colony Optimization (ACO), Machine Learning, Pattern Classification, Neural Networks.
Related
Ontology
Subjects/Areas/Topics:
Artificial Intelligence
;
Computational Intelligence
;
Evolutionary Computing
;
Knowledge Discovery and Information Retrieval
;
Knowledge-Based Systems
;
Machine Learning
;
Soft Computing
;
Swarm/Collective Intelligence
;
Symbolic Systems
Abstract:
Although artificial neural networks can be a very effective classification method, one of the drawbacks of their use is the need to manually prescribe the neural network topology. Recent work has introduced the ANN-Miner algorithm, an Ant Colony Optimization (ACO) technique for optimizing the topology of arbitrary FFNN's, i.e. FFNN's with multiple hidden layers, layer-skipping connections, and without the requirement of full-connectivity between successive layers. In this paper, we explore the use of several classification quality evaluation functions in ANN-Miner. Our experimental results, using 30 popular benchmark datasets, identify several quality functions that significantly improve on the simple Accuracy quality function that was
previously used in ANN-Miner.