Dendritic Cell Algorithm with Group Particle Swarm Optimization for Input Signal Generation

Abstract

Dendritic cell algorithm (DCA) is a classification algorithm that simulates the behavior of dendritic cells in the tissue environment. Selecting the most valuable attributes and assigning them a suitable signal categorization are crucial for DCA to generate input signals on the data pre-processing and initialization phase. Several methods were employed (e.g., Correlation Coefficient and Rough Set Theory). Those studies preferred to measure the importance of features based on the degree of relevance with the class and determined a mapping relationship between important features and signal categories of DCA based on expert knowledge. Generally, those researches ignore the effect of unimportant features, and the mapping relationship determined by expertise may not produce an optimal classification result. Thus, a hybrid model, GPSO-DCA, is proposed to accomplish feature selection and signal categorization based on Grouping Particle Swarm Optimization (GPSO) without any expertise. This study transforms feature selection and signal categorization into a grouping task (i.e., the selected features are divided into different signal groups) by redefining the data coding and velocity updating equations. The GPSO-DCA searches the optimal feature grouping scheme automatically instead of performing feature selection and signal categorization. The proposed approach is verified by employing the UCI Machine Learning Repository with significant performance improvement.