Implements the CAIRAD techique for detecting noisy values in a dataset. Does this with an analysis of coappearance between values. Can output whether or not a record is noisy (i.e. includes 1 or more noisy values), or remove all noisy values and replace them with missing values.

CAIRAD specification from:

Rahman, M. G., Islam, M. Z., Bossomaier, T., & Gao, J. (2012, June). Cairad: A co-appearance based analysis for incorrect records and attribute-values detection. In The 2012 International Joint Conference on Neural Networks (IJCNN) (pp. 1-10). IEEE. Available at http://doi.org/10.1109/ijcnn.2012.6252669

For more information, please see Associate Professor Zahid Islam's website here

@inproceedings{rahman2012cairad,
 author = {Rahman, Md Geaur, Islam, Md Zahidul, Bossomaier, Terry, and Gao, Junbin},
 title = {CAIRAD: A Co-appearance based Analysis for Incorrect Records and Attribute-values Detection},
 booktitle = {Proceedings of IEEE International Joint Conference on Neural Networks (IJCNN 12)},
 date = {10-15 June}
 year = {2012},
 isbn = {978-1-4673-1488-6},
 doi = {10.1109/IJCNN.2012.6252669}
 location = {Brisbane, QLD, Australia},
 pages = {2190--2199},
 url ={https://ieeexplore.ieee.org/abstract/document/6252669},
 publisher = {IEEE},
 keywords = {data pre-processing, data cleansing, data mining, noise detection},
}

Either download CAIRAD from the Weka package manager, or download the latest release from the "Releases" section on the sidebar of Github. A video showing the installation and use of the package can be found here

This repository houses a Netbeans project. Load the project into Netbeans to work on the package. Alternatively, download CAIRAD.java and import it into your Weka project to use it in your code.

-T coappearanceThreshold - Coappearance Threshold, tau in original paper.

-L coappearanceScoreThreshold - Coappearance Score Threshold, lambda in original paper.

-M makeNoisyMissing - Make detected noise into missing values.