I would be very grateful if you could direct me to publicly available dataset for clustering and/or classification with/without known class membership. Variables (attributes) should be continuous (no categorical/binary); and number of observation should be greater than number of variables. Number of variables less than 100 would be preferable.
Thanks
Michael
At the UCI Machine Learning Repository (http://archive.ics.uci.edu/ml/datasets.html) you can find over 300 data sets related to classification, clustering, regression and other ML tasks.
http://archive.ics.uci.edu/ml/datasets.html
Thanks Ales; I was aware of these dataset, and I got some but not all what I am looking for.
Have you seen following links?
https://sites.google.com/site/fawadsyed/datasets
http://mlg.ucd.ie/datasets/index.html#data
http://derekgreene.com/datasets/
If you are into text clustering/classification tasks then following link can be very useful
http://trec.nist.gov/
Is there any specific kind of data needed? Perhaps an specific domain? I'm working with data stream clustering and I have a few public datasets to share depending on your case.
Thanks Jean; we are developing a variable (feature) selection method for model-based clustering. As expected model-based clustering is quite computationally intensive, we are looking for reasonable number of continuous variables at the moment to test the method.
Thanks
Michael
Dear Michael,
How much is reasonable for you?
Take a look at Kaggle, especially in this dataset: https://www.kaggle.com/c/higgs-boson
Best Regards,
Jean
Thanks Jean for that; up to 3000 observations and 100 variables (features) would be enough at this stage. Thanks
UCI Machine Learning Datasets are available to all and are considered benchmarks
Thanks a lot Sebastian! That is great collection! At the mement am interested on continous variables, maximum about 100 variables, from cross-sectional study.
Thanks
Michael
Hi Sebastian; I posted this question in linkedin, link below, and a number of members suggested interesting dataset that might be addition to your collections although there are some overlap to those have been suggested here.
https://www.linkedin.com/groupItem?view=&gid=77616&type=member&item=5958592123753238532&commentID=5961760164078977024&report.success=8ULbKyXO6NDvmoK7o030UNOYGZKrvdhBhypZ_w8EpQrrQI-BBjkmxwkEOwBjLE28YyDIxcyEO7_TA_giuRN#commentID_5961760164078977024
OpenML.org has thousands of (mostly classification) datasets. All in the same format and downloadable via APIs. It also has all models built on those datasets.
Kaggle? They have a lot of great data. Titanic dataset is fun, they also have a pretty descent Imbd scrape. I use them sometimes in my analytics class.
A benchmark of datasets for cluster analysis every algorithm should be able to solve:
https://www.uni-marburg.de/fb12/arbeitsgruppen/datenbionik/Daten
There are lots of manually labeled text and image classification datasets are available here :
https://dataturks.com/projects/trending
Is it possible to get a data set of complete blood test and corresponding diseases found?
The set referred to by @Pasi Franti have served me a great deal. Several studies published in quality journals, have also featured the set.
There are groups of synthetic datasets in which one or two data parameters (size, dimensions, cluster variance, overlap, etc) are varied across the member datasets, to help study how an algorithm performs as such condition(s) are varied.
The site also lists several real-world datasets.
http://cs.uef.fi/sipu/datasets/
- Badges
- Science topic
- Similar topics
- Data Mining
- Classification