Introduction

An Example

The Algorithm

Constructive Induction

Related Work

Our Papers on Function Decompositon



Orange

Data sets

Spotlight

Looking for a public data mining software? Try Orange, a scriptable component based framework.

Want to use decision support tool, but your computer is just not there when you need it? Check our handheld-based decision-support schema and our PalmPilot software.

Explore what we are doing in functional genomics.

FRI > Biolab > Function Decomposition > Our Papers on Function Decomposition

Our papers

If you are new to function decomposition, start with the paper that appeared in IEEE Intelligent Systems and their Applications [2]. It describes the basic decomposition algorithm that deals with multi-valued attributes and class, and discovers classifiers that are 100% consistent with original data sets. We named this algorithm as minimal-complexity decomposition, as the method thrives to decomposition the original data set to a set of smaller data sets. An extended version of this paper appeared in the book by H. Liu and H. Motoda [3].

Minimal-complexity decomposition is in the detail described in the paper that appeared in Artificial Intelligence Journal [5]. The shorter version of this paper was presented in ICML-97 [4]. Different partition selection methods are investigated in the paper published in Informatica [11].

A paper at KDD-97 [6] presents how function decomposition can be viewed within a data mining framework. It is in this paper that we first introduce the so-called "supervised" function decomposition, that is, a function decomposition that is to guided by the human expert in selecting the concepts that are worth constructing. The idea was pushed further by allowing the expert to, if needed, change or suggest some parts of the functions that are discovered by decomposition [10], but further investigations (and a program with a nice user interface) is needed. How function decomposition may be considered as a tool to automatically construct a decision support models was presented at ISDSS-1997 [7].

Noise handling in function decomposition is one of the main contributions of the PhD Thesis by B. Zupan [1]. This part of the Thesis was subsequently presented at ICML-2000 [8].

Function decomposition and its application in handling continuous data was only marginally studied. One of the few attempts to deal with this problem was presented at ECML-97 [9].

[Disclaimer: We are making the pre-prints of the papers available in PDF. To see the printed version of the papers, please check original publication.]

[1] Zupan B: Machine learning based on function decomposition. PhD Thesis, University of Ljubljana, 1997.

[2] Zupan B, Bohanec M, Demsar J, Bratko I: Feature transformation by function decomposition . IEEE Intelligent Systems & Their Applications, 1998, vol. 13, pages 38-43.

[3] Zupan B, Bohanec M, Demsar J, Bratko I: Feature transformation by function decomposition, In: Feature extraction, construction and selection: a data mining perspective (Liu H, Motoda H, eds.), Kluwer, Boston, 1998, pages 325-340.

[4] Zupan B, Bohanec M, Bratko I, Demsar J: Machine learning by function decomposition, In: Machine Learning: Proceedings of the Fourteenth International Conference (ICML'97), Nashville, Tennessee, July 1997, pages 421-429.

[5] Zupan B, Bohanec M, Demsar J, Bratko I: Learning by discovering concept hierarchies. Artificial Intelligence, 1999, vol. 109, pages 211-242.

[6] Zupan B, Bohanec M, Bratko I, Cestnik B: A dataset decomposition approach to data mining and machine discovery. In: Proceedings of the Third International Conference on Knowledge Discovery and Data Mining, Newport Beach, CA, August 1997, pages 299-302.

[7] Bohanec M, Zupan B, Bratko I, Cestnik B: A function-decomposition method for development of hierarchical multi-attribute decision models, In: Proceedings of the Fourth Conference of the International Society for Decision Support Systems: Lausanne, Switzerland, July 1997, pages 503-514.

[8] Zupan B, Bratko I, Bohanec M, Demsar J: Induction of concept hierarchies from noisy data, In: Proceedings of the Seventeenth International Conference on Machine Learning, (ICML-200): June 2000, San Francisco, CA, 2000, pages 1199-1206.

[9] Demsar J, Zupan B, Bohanec M, Bratko I: Constructing intermediate concepts by decomposition of real functions, In: Proceedings of the 9th European Conference on Machine Learning, Prague, Czech Republic, April 1997, pages 93-107.

[10] Krisper M, Zupan B: Synthesis of hierarchical decision support models from socioeconomic data, In: Procedeeings of the Information Society Conference, Ljubljana, Slovenia, 1998, pages 60-63.

[11] Zupan B, Bohanec M. Experimental evaluation of three partition selection criteria for decision table decomposition. Informatica, 1998, vol. 22, pages 207-217.