Software
There are some software developed in my research. Those software are provided here along with quick instructions in how to use them.
Feature Selection
A feature selection extension (plug-in) for RapidMiner 5 is available for downloading (FeatureSelectionEx.jar). The extension currently includes the implementations of the following feature selection methods:
- KLS-FS: a unsupervised feature selection method using incremental least squares
- SDJ-FS: a supervised feature selection method based on fractal dimension (spatial distance join)
- Download the extension (FeatureSelectionEx.jar) if you have not done so,
- Save the downloaded JAR file to folder lib\plugins under the root director of your rapidminer,
- Open RapidMiner and you will find the feature selection methods located in a new folder named Feature Selection Extension under the Operators Tab (see box 1 in the figure below).
To use KLS-FS feature selection method:
If you want to refer to the KLS-FS and SDJ-FS feature selection methods, the full citations are
- Drag the KLSFS operator (box 1 in above figure) in the main process frame and connect its input (exa) to a data source,
- Click on the KLSFS operator (box 2) to bring up the parameter setting frame (box 3) and the help frame (box 4 about how to use KLSFS),
- Set proper parameters according to the description given in the help frame (note that the exa out of KLSFS provides only the original data, while calculated feature weights are provided by wei output. The weight of a feature calculated by KLSFS is actually its Relative Least Square Error and zero weight indicates the feature is not selected),
- Connect the two outputs of KLSFS to a Select by Weights operator (box 5, which is usually used in pair with a feature selection operator to select features according to feature weights calculated by the feature selection operator),
- Choose "greater" as the "weight relation" parameter of the Select by Weights operator and set the "weight" parameter as zero (this will select only the features of weight greater than zero),
- Port the selected features provided by the exa output of the Select by Weights operator to other operators for model development based on the selected features.
If you want to refer to the KLS-FS and SDJ-FS feature selection methods, the full citations are
- R. Liu, R. Rallo, and Y. Cohen, Unsupervised Feature Selection Using Incremental Least Squares. International Journal of Information Technology & Decision Making, 2011. 10(6): 967-987.
- R. Liu, Y. Shi, Spatial distance join based feature selection, Engineering Applications of Artificial Intelligence, 2013. 26(10):2597-2607