Computing and Informatics

Feature selection (FS) averts the consideration of unwanted features which may tend the classification algorithm to classify wrongly. Choosing an optimal feature subset from the given set of features is challenging due to the complex associations present within the features. In non-convex conditions, the gradient-based algorithms suffer due to local optima or saddle points with respect to initial conditions where swarm intelligence algorithms pose a higher chance to converge over the global optima. The Salp Swarm Algorithm (SSA) proposed by Mirjalili et al. is based on the chaining behaviour of sea salps but the algorithm lacks diversity in the exploration stage. Rectifying the exploratory behaviour and testing the algorithm against the FS problem is the motivation behind this work. Three variants of the algorithm are proposed, of which the Vigilant Salp Swarm Algorithm (VSSA) inherits the vigilant mechanism in Grey Wolf Optimizer (GWO), the second variant and the third variant replace a simple crossover operator and shuffle crossover operator instead of the follower's position update mechanism used in the VSSA to form Vanilla Crossover VSSA (VCVSSA) and Shuffle Crossover VSSA (SCVSSA).