LIBSVM Data: Classification, Regression, and Multi-label

This page contains many classification, regression, multi-label and string data sets stored in LIBSVM format. For some sets raw materials (e.g., original texts) are also available. These data sets are from UCI, Statlog, StatLib and other collections. We thank their efforts. For most sets, we linearly scale each attribute to [-1,1] or [0,1]. The testing data (if provided) is adjusted accordingly. Some training data are further separated to "training" (tr) and "validation" (val) sets. Details can be found in the description of each data set. To read data via MATLAB, you can use "libsvmread" in LIBSVM package.

A summary of all data sets is in the following. If you have used LIBSVM with these sets, and find them useful, please cite our work as:
Chih-Chung Chang and Chih-Jen Lin, LIBSVM : a library for support vector machines. ACM Transactions on Intelligent Systems and Technology, 2:27:1--27:27, 2011. Software available at http://www.csie.ntu.edu.tw/~cjlin/libsvm.

Please also cite the source of the data sets (references given below).

Go to pages of classification (binary, multi-class), regression, multi-label, and string. Those interested in hierarchical data with many classes can visit LSHTC page.

Some sets are large and the connection may fail. On Linux you can use

> wget -t inf URL_address_of_data

to retry infinitely many times. If it still fails, add -c to continuely get a partially-downloaded set. You can also use

> lftp -c 'pget -c URL_address_of_data'

to have several connections for reducing the downloading time.

name	source	type	class	training size	testing size	feature
a1a	UCI	classification	2	1,605	30,956	123
a2a	UCI	classification	2	2,265	30,296	123
a3a	UCI	classification	2	3,185	29,376	123
a4a	UCI	classification	2	4,781	27,780	123
a5a	UCI	classification	2	6,414	26,147	123
a6a	UCI	classification	2	11,220	21,341	123
a7a	UCI	classification	2	16,100	16,461	123
a8a	UCI	classification	2	22,696	9,865	123
a9a	UCI	classification	2	32,561	16,281	123
australian	Statlog	classification	2	690		14
avazu	Avazu's Click-through Prediction	classification	2	40,428,967	4,577,464	1,000,000
breast-cancer	UCI	classification	2	683		10
cod-rna	[AVU06a]	classification	2	59,535		8
colon-cancer	[AU99a]	classification	2	62		2,000
covtype.binary	UCI	classification	2	581,012		54
criteo	Criteo's Display Advertising Challenge	classification	2	45,840,617	6,042,135	1,000,000
criteo_tb	Criteo's Terabyte Click Logs	classification	2	4,195,197,692	178,274,637	1,000,000
diabetes	UCI	classification	2	768		8
duke breast-cancer	[MW01a]	classification	2	44		7,129
epsilon	PASCAL Challenge 2008	classification	2	400,000	100,000	2,000
fourclass	[TKH96a]	classification	2	862		2
german.numer	Statlog	classification	2	1,000		24
gisette	NIPS 2003 Feature Selection Challenge [IG05a]	classification	2	6,000	1,000	5,000
heart	Statlog	classification	2	270		13
HIGGS	UCI	classification	2	11,000,000		28
Hyperpartisan News Detection	SemEval-2019 Task 4: Hyperpartisan News Detection	classification	2	516	65
ijcnn1	[DP01a]	classification	2	49,990	91,701	22
imdb-sentiment	Learning Word Vectors for Sentiment Analysis	classification	2	25,000	25,000
ionosphere	UCI	classification	2	351		34
kdd2010 (algebra)	KDD CUP 2010	classification	2	8,407,752	510,302	20,216,830
kdd2010 (bridge to algebra)	KDD CUP 2010	classification	2	19,264,097	748,401	29,890,095
kdd2010 raw version (bridge to algebra)	KDD CUP 2010	classification	2	19,264,097	748,401	1,163,024
kdd2012	KDD CUP 2012	classification	2	149,639,105		54,686,452
leukemia	[TG99a]	classification	2	38	34	7129
liver-disorders	UCI	classification	2	145	200	5
madelon	NIPS 2003 Feature Selection Challenge [IG05a]	classification	2	2,000	600	500
mushrooms	UCI	classification	2	8124		112
news20.binary	[SSK05a]	classification	2	19,996		1,355,191
phishing	UCI	classification	2	11,055		68
rcv1.binary	[DL04b]	classification	2	20,242	677,399	47,236
real-sim	A. McCallum	classification	2	72,309		20,958
skin_nonskin	UCI	classification	2	245,057		3
splice	Delve	classification	2	1,000	2,175	60
splice-site	[SS10a,AA12a]	classification	2	50,000,000	4,627,840	11,725,480
sonar	UCI	classification	2	208		60
SUSY	UCI	classification	2	5,000,000		18
svmguide1	[CWH03a]	classification	2	3,089	4,000	4
svmguide3	[CWH03a]	classification	2	1,243	41	21
url	[JM09a]	classification	2	2,396,130		3,231,961
w1a	[JP98a]	classification	2	2,477	47,272	300
w2a	[JP98a]	classification	2	3,470	46,279	300
w3a	[JP98a]	classification	2	4,912	44,837	300
w4a	[JP98a]	classification	2	7,366	42,383	300
w5a	[JP98a]	classification	2	9,888	39,861	300
w6a	[JP98a]	classification	2	17,188	32,561	300
w7a	[JP98a]	classification	2	24,692	25,057	300
w8a	[JP98a]	classification	2	49,749	14,951	300
webspam	Webb Spam Corpus [ST06a]	classification	2	350,000		16,609,143
aloi	aloi [AR14a]	classification	1,000	108,000		128
cifar10	The CIFAR-10 dataset [AK09a]	classification	10	50,000	10,000	3,072
connect-4	UCI	classification	3	67,557		126
covtype	UCI	classification	7	581,012		54
dna	Statlog	classification	3	2,000	1,186	180
glass	UCI	classification	6	214		9
imdb-rating	Jointly Modelling Aspects, Ratings and Sentiments for Movie Recommendation	classification	10	348,415
iris	UCI	classification	3	150		4
LEDGAR (LexGLUE)	[IC22b]	classification	100	60,000	10,000	19,996
letter	Statlog	classification	26	15,000	5,000	16
mnist	[YL98a]	classification	10	60,000	10,000	780
mnist8m	Invariant SVM [GL07b]	classification	10	8,100,000		784
news20	[KL95a]	classification	20	15,935	3,993	62,061
news20 (18,846)	[KL95a]	classification	20	9,051	7,532	130,107
pendigits	UCI	classification	10	7,494	3,498	16
poker	UCI	classification	10	25,010	1,000,000	10
protein	[JYW02a]	classification	3	17,766	6,621	357
rcv1.multiclass	[DL04b]	classification	53	15,564	518,571	47,236
SCOTUS (LexGLUE)	[IC22b]	classification	13	5,000	1,400	126,405
satimage	Statlog	classification	6	4,435	2,000	36
sector	[AM98a]	classification	105	6,412	3,207	55,197
segment	Statlog	classification	7	2,310		19
Sensorless	UCI	classification	11	58,509		48
shuttle	Statlog	classification	7	43,500	14,500	9
smallNORB	The Small NORB Dataset [YL04b]	classification	5	24,300	24,300	18,432
SVHN	SVHN [YN11a]	classification	10	73,257	26,032	3,072
svmguide2	[CWH03a]	classification	3	391		20
svmguide4	[CWH03a]	classification	6	300	312	10
usps	[JJH94a]	classification	10	7,291	2,007	256
SensIT Vehicle (acoustic)	Sensit [MD04a]	classification	3	78,823	19,705	50
SensIT Vehicle (seismic)	Sensit [MD04a]	classification	3	78,823	19,705	50
SensIT Vehicle (combined)	Sensit [MD04a]	classification	3	78,823	19,705	100
vehicle	Statlog	classification	4	846		18
vowel	UCI	classification	11	528	462	10
wine	UCI	classification	3	178		13
abalone	UCI	regression		4,177		8
bodyfat	StatLib	regression		252		14
cadata	StatLib	regression		20,640		8
cpusmall	Delve	regression		8,192		12
E2006-log1p	10-K Corpus	regression		16,087	3,308	4,272,227
E2006-tfidf	10-K Corpus	regression		16,087	3,308	150,360
eunite2001		regression		336	31	16
housing	UCI	regression		506		13
mg	[GWF01a]	regression		1,385		6
mpg	UCI	regression		392		7
pyrim	UCI	regression		74		27
space_ga	StatLib	regression		3,107		6
triazines	UCI	regression		186		60
YearPredictionMSD	UCI	regression		463,715	51,630	90
Amazon-670K	[JM13a]	multi-label	670,091	490,449	153,025	135,909(ver1) 135,909(ver2)
AmazonCat-13K	[JM13a]	multi-label	13,330	1,186,239	306,782	203,882 (ver1) 1,293,747 (ver2) 203,882(ver3)
bibtex	[GT08a]	multi-label	159	7,395		1,836
BlogCatalog	[LT09a]	multi-label	39	10,312		128
delicious	[GT08a]	multi-label	983	16,105		500
ECtHR (A) (LexGLUE)	[IC22b]	multi-label	10	9,000	1,000	92,401
ECtHR (B) (LexGLUE)	[IC22b]	multi-label	10	9,000	1,000	92,401
EUR-LEX (LexGLUE)	[IC22b]	multi-label	100	55,000	5,000	147,464
EUR-Lex	[LM10a]	multi-label	3,956	15,449	3,865	186,104
EURLEX57K	[IC19a]	multi-label	4,271	45,000	6,000	N/A
Flickr	[LT09a]	multi-label	195	80,513		128
mediamill (exp1)	Mediamill	multi-label	101	30,993	12,914	120
PPI	[WLH17a]	multi-label	121	54,958		128
rcv1v2 (topics; subsets)	[DL04b]	multi-label	101	3,000	3,000	47,236
rcv1v2 (topics; full sets)	[DL04b]	multi-label	101	23,149	781,265	47,236
rcv1v2 (industries; full sets)	[DL04b]	multi-label	313	23,149	781,265	47,236
rcv1v2 (regions; full sets)	[DL04b]	multi-label	228	23,149	781,265	47,236
scene-classification	[MB04a]	multi-label	6	1,211	1,196	294
siam-competition2007	SIAM Text Mining Competition 2007	multi-label	22	21,519	7,077	30,438
UNFAIR-ToS (LexGLUE)	[IC22b]	multi-label	8	5,532	1,607	6,290
Wiki10-31K	[AZ09a]	multi-label	30,938	14,146	6,616	104,374
yeast	[AE02a]	multi-label	14	1,500	917	103
mnist (string format)	[SL96a]	string		60,000	10,000	string
YouTube	[LT09a]	multi-label	46	31,703		128

We have tried the best to obtain the permission from most original sources for distributing these sets. Please follow their respective copyrights for using them.

Author: Rong-En Fan at National Taiwan University. Please contact Chih-Jen Lin for any question.