A bacterial evolutionary algorithm for feature selection

• Mario Drobics
• János Botzheim

When creating regression models from data the problem arises that the complexity of the models rapidly increases with the number of features involved. Especially in real world application where a large number of potential features are available, feature selection becomes a crucial task. A novel approach for feature selection is presented which uses a bacterial evolutionary algorithm to identify the optimal set and the optimal number of features with respect to a given learning problem and a given learning algorithm. This method ensures high accuracy and significantly increases interpretability of the resulting models.