[ R-value ]

1. Introduction
 The quality of dataset has a profound effect on classification accuracy, and there is a clear need for some method to evaluate this quality. R-value is a new dataset evaluation method. This proposed method is based on the ratio of overlapping areas among categories in a dataset. A high R-value for a dataset indicates that the dataset contains wide overlapping areas among its categories (classes), and classification accuracy on the dataset may become low. We can use the R-value measure to understand the characteristics of a dataset, the feature selection process, and the proper design of new classifiers. R-value captures overlapping areas as shown in Figure 1.


Figure1. Concept of R-value


** 3 types of R-value

  • R-value between two categories (Figure 2(a))
  • R-value of a category (Figure 2(b))
  • R-value of a whole dataset (Figure 2(c))


Figure2. Three types of R-valuez

2. Usage
 Format of Excution:

java calcRvalue [dataset file name] [K value] [seta value]

Options:

-dataset file name : dataset file for evaluation
-K value : number of nearest neighbors (optional)
-seta value : threshold to decide if a sample belongs to overlapping area or not (optional)

Examples

java calcRvalue mydataset.csv
java calcRvalue mydataset.csv 7 3

3. Download

More sample Dataset

Note
  1. Dataset should be .csv format.
  2. Only numerical vaule is allowed for dataset.
  3. First column of dataset should be class(category) data and shold have continuous value beginning with 0.(0,1,2,3,...)

4. Citation Request:
 Sejong Oh, Improved Measures of Redundancy and Relevance for mRMR Feature Selection.