Ant colony classification mining algorithm based on pheromone attraction and exclusion

• Published in: Soft computing (Berlin, Germany) Vol. 21; no. 19; pp. 5741 - 5753
• Main Authors: Yang, Lei, Li, Kangshun, Zhang, Wensheng, Ke, Zhenxu
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.10.2017; Springer Nature B.V
• Subjects: Accuracy; Algorithms; Ant colony optimization; Artificial Intelligence; Classification; Computational Intelligence; Control; Data mining; Datasets; Decision trees; Engineering; Food; Foraging behavior; Genetic algorithms; Heuristic; Iterative methods; Machine learning; Mathematical Logic and Foundations; Mechatronics; Methodologies and Application; Neural networks; Optimization; Pattern recognition; Pheromones; Robotics; Rule induction; Search process; Classification rule; Data mining; Ant colony algorithm; Pheromone attraction and exclusion
• ISSN: 1432-7643; 1433-7479
• DOI: 10.1007/s00500-016-2151-9

Ant colony optimization algorithms have been applied successfully in classification rule mining. However, basic ant colony classification mining algorithms generally suffer from problems, such as premature convergence and falling into local optimum easily. Simultaneously, the classification mining algorithms use sequential covering strategy to discover rules, and the interaction between rules is less considered. In this study, a new ant colony classification mining algorithm based on pheromone attraction and exclusion (Ant-Miner PAE ) is proposed, in which a new pheromone calculation method is designed and the search is guided by the new probability transfer formula. By contrast,the basic algorithm structure is modified, and the order of the iteration is adjusted. Thus, the problem of rule interaction is mitigated. Ant-Miner PAE can balance the relation of exploration and development of constructing rules, which can make the ants in the search process initially explore and develop in the later period. Our experiments, which use 12 publicly available data sets, show that the predictive accuracy obtained by Ant-Miner PAE implementing the proposed pheromone attraction and exclusion strategy is statistically significantly higher than the predictive accuracy of other rule induction classification algorithms, such as CN2, C4.5 rules, PSO/ACO2, Ant-Miner, and c Ant-Miner PB . Furthermore, the rules discovered by Ant-Miner PAE are considerably simpler than those discovered by its counterparts.