An immune-inspired online feature selection mechanism

A solution to the Feature Selection Problem (FSP) still remains one of the challenges faced in the domain of machine learning. Most of the solutions to the FSP are designed and proposed as a preprocessing step wherein the best suited features are selected in an off-line manner. Online feature selection is a much more complex problem. This paper describes a novel Immuno-inspired Online Feature Selection (iFS) approach to select the best suited features for a domain. The approach uses an Immune Network as a base to realize an autonomous and self-learning solution to online FSP. Unlike conventional methods this approach does not rely on a fixed set of predefined training data. Instead, given a set of features (antibodies) it responds to incoming data (antigens) and delivers a reduced set of the former. The efficacy of the proposed method has been justified based on experiments carried out to classify human beings from non-human ones. Comparisons with three widely used methods also authenticate the superiority of the approach.