Data-mining tools for time-dependent, mixed-type attribute database, including feature construction

Tools: Checkup, FEADIS, VECDIS

Ofer Dor, Yoram Reich


We are developing a method that creates new features through evolution. Our extensive experiments show that the method can boost up the performance of almost all classifiers on almost all datasets. The amount of performance improvement observed has been significant.

We also demonstrate a phenomenon that by constructing new features one can take almost any classifier and boost its performance to the best known performance today n that database.



Due to these results, which are based totally on empirical algorithm without sophisticated equations, we encountered people who were skeptical about them. In order to allow skeptics test the results themselves and allow the community at large to experience this performance boost, we provide the program and the data used in our papers for download at the bottom of this page.

If you make use of them, please refer to the relevant paper. We would like to hear how the method worked for you.


Keywords: features discovery, feature generation, constructive induction, feature selection, time series, stacked generalization, evolutionary programming, ensemble, no-free lunch theorem, free lunch


1.    O. Dor, Y. Reich, CHECKUP: A feature generation rule-base learner, The Israeli Association for Artificial Intelligence, IAAI07 Symposium, Ashkelon College, 2007.

2.    O. Dor, Y. Reich, An evaluation of musical score characteristics for automatic classification of composers, Computer Music Journal, 35(3):86-97, 2011.

3.    O. Dor, Y. Reich, Strengthening learning algorithms by feature discovery, Information Sciences, 189:176-190, 2012.

4.    O. Dor, Y. Reich, Enhancing learning algorithms to support data with short sequence features by automated feature discovery, Knowledge-Based Systems, 52:114-132, 2013.


1.      db_composers database for the CMJ journal. If you make use of this database, please cite the Computer Music Journal paper above.

2.      FEADIS FEAture DIScovery system (updated 8/12/11). If you make use of this software and/or data, please cite the Information Sciences paper above.

3.      VECDIS VECtor DIScovery system (updated 9/8/13). If you make use of this software and/or data, please cite the Knowledge-Based Systems paper above.


Copyright 2005-14 Yoram Reich
Page URL:

Last modified: 4/7/2014 2:53:00 PM