Discovering Unordered Rule Sets for Mixed Variables Using Ant Colony Optimization

Ant colony optimization (ACO) algorithms have been applied successfully to combinatorial optimization problems. Ant colony optimization (ACO) can be applied to the data mining field to extract rule-based classifier. The Ant-Miner algorithm was first proposed by Parpinelli and his colleagues (2002), applies an Ant Colony Optimization (ACO) heuristic to the classification task of data mining to discover an ordered list of classification rules. Ant-Miner uses a discretization process to deal with continuous attributes in the data set. The transition rule and pheromone update strategy are the more problem independent features of Ant-Miner algorithm. In Ant–Miner algorithm a term had stochastically chosen and added to the current partial rule. This paper presents an enhanced Ant Miner, which includes two main contributions. Firstly, use Laplace-corrected confidence heuristic function to generate unordered rule sets for mixed variables and combine both stochastic selection and deterministic selection to include a term in the current partial rule to improve the predictive accuracy of a discovered rule. Secondly, it handles both continuous and discrete attributes without discretization using multimodal functions and produces a dynamic balance between diversification and intensification during search process. The result of the algorithm had compared against the original Ant-Miner algorithm in six public-domain datasets with respect to predictive accuracy and simplicity of the discovered rules. Empirical results show that the enhanced Ant-Miner can quickly discover better classification rules, which have roughly competitive predicative accuracy, and short rules.