Ioannis Michelakos

Abstract:

Abstract:

Abstract: This paper presents a study of the user-defined parameters of the cAnt-Miner2 algorithm which extended the approach of coping with continuous attributes that was introduced by the cAnt-Miner algorithm. Experiments on two public medical data sets were carried out, in order to find appropriate parameter combinations for different classification problems. The experiments highlight the influence of each parameter and their combinations on the cAnt-Miner2 algorithm, in terms of accuracy, simplicity and computational cost.

Abstract: Ant colony optimization algorithms have been applied successfully to data mining classification problems. Recently, an improved version of cAnt-Miner (Ant-Miner coping with continuous attributes), called cAnt-Miner2, has been introduced for mining classification rules. In this paper, a hybrid algorithm is presented, combining the cAnt-Miner2 and the mRMR feature selection algorithms. The proposed algorithm was experimentally compared to cAnt-Miner2, using some public medical data sets to demonstrate its functioning. The experiments were very promising and the proposed approach is better in terms of accuracy, simplicity and computational cost than the original cAnt-Miner2 algorithm.