Aggregation pheromone metaphor for semi-supervised classification

This article presents a novel ‘self-training’ based semi-supervised classification algorithm using the property of aggregation pheromone found in real ants. The proposed method has no assumption regarding the data distribution and is free from parameters to be set by the user. It can also capture arbitrary shapes of the classes. The proposed algorithm is evaluated with a number of synthetic as well as real life benchmark datasets in terms of accuracy, macro and micro averaged F1 measures. Results are compared with two supervised and three semi-supervised classification techniques and are statistically validated using paired t-test. Experimental results show the potentiality of the proposed algorithm.

► Proposed novel ant based semi-supervised classification algorithm.
► The algorithm is self-training in nature.
► Proposed method is parameter free with no assumption regarding data distribution.
► Method can adaptively capture arbitrary shapes of classes.
► Potentiality of the method is justified from the experimental results.