A Swarm Intelligence Approach in Undersampling Majority Class

Over the years, machine learning has been facing the issue of imbalance dataset. It occurs when the number of instances in one class significantly outnumbers the instances in the other class. This study investigates a new approach for balancing the dataset using a swarm intelligence technique, Stochastic Diffusion Search (SDS), to undersample the majority class on a direct marketing dataset. The outcome of the novel application of this swarm intelligence algorithm demonstrates promising results which encourage the possibility of undersampling a majority class by removing redundant data whist protecting the useful data in the dataset. This paper details the behaviour of the proposed algorithm in dealing with this problem and investigates the results which are contrasted against other techniques.