Benchmarks

The PNC2 cluster algorithm was developed and compared to the most similiar existing algrithms using several freely available benchmarks. Take a look at my PhD thesis if you want to see the results. Below you find a short desciption of the benchmarks used and you may download them in a format directly readable to the PNC2 Rule Induction System. Most of the benchmarks originate from the UCI Machine Learning Repository. Go there if you want the original format/version or any additional info.

Overview and Download

Benchmark	Output column	Classes	Tuples	Inputs	nominal / continous	Missing features	Origin
Australian	15	2	690	14	8/6	No	UCI
Cleveland	14	2	303	13	9/4	Yes	UCI
Chess-End-Game	37	2	3196	36	36/0	No	UCI
DNA	61	3	3175	60	0/60	No	Delve
Hepatitis	1	2	155	19	13/6	Yes	UCI
House-Votes-84	1	2	435	16	16/0	Yes	UCI
Iris	5	3	150	4	0/4	No	UCI
Letter-Recognition	1	26	20000	16	0/16	No	UCI
Mushrooms	1	2	8124	22	22/0	Yes	UCI
Segmentation	1	7	2130	19	0/19	No	UCI
Waveform	22	3	5000	21	0/21	No	UCI
Waveform+Noise	41	3	5000	40	0/40	No	UCI
Wine	1	3	178	13	0/13	No	UCI

Note: The data sample DNA used by Delve was build from the one of the UCI repository by removing 15 data tuples.

Short Description

Australian Verification of credit card transactions based on anonymous customer information. The original data contained about 5% of data tuples with one or more missing feature values. These were substituted by the mean for continuous and by the most frequent symbol for nominal features.
Chess-End-Game Decide if player white is able to win the game based upon a given situation on a chess board. Also known as King+Rook versus King+Pawn on a7 (KRKPA7)
Cleveland Decide if a patient suffers from a heart disease based upon the patient's record. The original output took integer values from 0 to 4, whereas 0 means, that the patient is healthy and 1 to 4 indicates, that the patient suffers with increasing degree from a heart disease.
DNA Classify splice junctions in DNA sequences. Decide upon a given window of 60 nucleoid sequences, if there is an intron-exon, and exon-intron or no boundary in the middle of the window.
Hepatitis Decide if a patient suffers from hepatitis based upon the patient's record.
House-Votes-84 Classify the party of a congress member based upon his votes to a couple of questions.
Iris Classify iris flowers based upon sepal and petal length and width.
Letter-Recognition Classify letters displayed in 20 different fonts based on some calculated features.
Mushrooms Classify if a mushroom is definitely edible, definitely poisonous, or of unknown edibility. The latter case counts as poisonous.
Segmentation Classify the type of surface for a pixel in a satellite image.
Waveform Classify the two out of three possible waveforms, that were used to build a measured signal.
Waveform+Noise The same as Waveform, but there are 19 additional inputs which contain nothing but random white noise.
Wine Classify a wine's cultivar based on a chemical analysis.