IEEE Neural Networks Council
Standards Committee

Working Group on Data Modeling Benchmarks

Goal

The goal of the Working Group on Data Modeling Benchmarks (under IEEE Neural Networks Council Standards Committee) is to provide an easy access to references to modeling approaches and related datasets via WWW. Hopefully this can facilitate further research and comparisons on computational data modeling approaches, including artificial neural networks, fuzzy inference systems, CART, MARS, and all nonlinear regression and optimization techniques.

This page is constantly under construction, but the lists of datasets/publications are by no means complete. If you have new datasets/papers and would like to put them here, please send me an email (jang@cs.nthu.edu.tw) to let me know where to link them. Any feedbacks and suggestions are also highly welcome.

J.-S. Roger Jang
CS Dept., Tsing Hua Univ., Taiwan

Data Sets

Mackey-Glass Chaotic Time Series

Raw dataset for tau = 17 and t = 0 to 1200.
C program that generates the raw dataset
Related Publication: Jan93itsmc, Chi94

Automobile MPG Database (from UCI Machine Learning Repository)

Revised from CMU StatLib library
Data concerns city-cycle fuel consumption
Continuously valued class attribute (mpg)
398 instances, 5 numeric attributes
FTP access
Related Publication: Jan96fuzz

A Nonlinear Function

50 instances, 4 numeric attributes (2 are dummy, for testing input selection). The data is generated from a nonlinear function output = (1+x^{-2}+y^{-1.5})^2.
(The data set was obtained from Table II of SugY93. However, the 42-nd tuple of the data set has a output value of "1.97", which is likely to be a outlier since the desired value, according to the above equation, should be "3.11".)
Related Publication: LinC95, SugY93

Box-Jenkins Furnace Data

Data set and description
Related Publication: Jan96fuzz, Chi96, SugY93

Human Operation at a Chemical Plant

70 instances, 5 numeric attributes
Related Publication: SugY93

Daily Price of Stock

100 instances, 10 numeric attributes
Related Publication: SugY93

Publications

(In no particular order)

@ARTICLE{LinC95,
	AUTHOR = {Yinghua Lin and George A. Cunningham III},
	TITLE = {A new approach to fuzzy-neural system modeling},
	JOURNAL = {IEEE Trans. on Fuzzy Systems},
	YEAR = {1995},
	VOLUME = {3},
	NUMBER = {2},
	PAGES = {190-198},
	MONTH = may
}

@ARTICLE{SugY93,
	AUTHOR = {M. Sugeno and T. Yasukawa},
	TITLE = {A fuzzy-logic-based approach to qualitative modeling},
	JOURNAL = {IEEE Trans. on Fuzzy Systems},
	YEAR = {1993},
	VOLUME = {1},
	NUMBER = {1},
	PAGES = {7-31},
	MONTH = feb
}

@ARTICLE{Jan93itsmc,
	AUTHOR = {J.-S. Roger Jang},
	TITLE = {ANFIS: Adaptive-Network-Based Fuzzy Inference Systems},
	JOURNAL = {IEEE Trans. on System, Man, and Cybernetics},
	YEAR = {1993},
	VOLUME = {23},
	NUMBER = {03},
	PAGES = {665-685},
	MONTH = may
}

Compressed PS file
Related datasets: Mackey-Glass time series

@ARTICLE{Chi94,
	AUTHOR = {S. L. Chiu},
	TITLE = {Fuzzy model identification based on
		cluster estimation},
	JOURNAL = {Journal of Intelligent and Fuzzy Systems},
	YEAR = {1994},
	VOLUME = {2},
	NUMBER = {3}
}

@INPROCEEDINGS{Jan96fuzz,
	AUTHOR = {J.-S. Roger Jang},
	TITLE = {Input selection for {ANFIS} learning},
	BOOKTITLE = {Proceedings of IEEE International Conference on
		Fuzzy Systems},
	YEAR = {1996},
	ADDRESS = {New Orleans},
	MONTH = sep
}

Compressed PS file
Related datasets: Automobile MPG database, Box-Jenkins furnace data

@ARTICLE{Chi95,
	AUTHOR = {Stephen L. Chiu},
	TITLE = {Selecting input variables for fuzzy models},
	JOURNAL = {Journal of Intelligent and Fuzzy Systems},
	YEAR = {1996},
	NOTE = {Forthcoming}
}