Abstract: The increase of dimensionality of data is a target for many existing feature selection methods with respect to efficiency and effectiveness. In this paper, the all relevant feature selection method based on information gathered using generational feature selection is described. The successive generations of feature subset were defined using
Rule Quality Importance algorithm and next the subset of most important features was eliminated from the primary dataset. This process was executed until the most relevant feature has got importance value on the level equal to importance of the random, shadow feature. The proposed approach was also tested on well-known artificial Madelon dataset and the results confirm its efficiency. Thus, the conclusion is that the identified features are relevant but not all weakly relevant features were discovered.