[其他] 【独家发布】lung-cancer data [推广有奖]

1. Title: Lung Cancer Data

2. Source Information:
    - Data was published in :
      Hong, Z.Q. and Yang, J.Y. "Optimal Discriminant Plane for a Small
      Number of Samples and Design Method of Classifier on the Plane",
      Pattern Recognition, Vol. 24, No. 4, pp. 317-324, 1991.
    - Donor: Stefan Aeberhard, stefan@coral.cs.jcu.edu.au
    - Date : May, 1992

3. Past Usage:
    - Hong, Z.Q. and Yang, J.Y. "Optimal Discriminant Plane for a Small
          Number of Samples and Design Method of Classifier on the Plane",
          Pattern Recognition, Vol. 24, No. 4, pp. 317-324, 1991.
    - Aeberhard, S., Coomans, D, De Vel, O. "Comparisons of
      Classification Methods in High Dimensional Settings",
      submitted to Technometrics.
    - Aeberhard, S., Coomans, D, De Vel, O. "The Dangers of
      Bias in High Dimensional Settings", submitted to
      pattern Recognition.

4. Relevant Information:
    - This data was used by Hong and Young to illustrate the
      power of the optimal discriminant plane even in ill-posed
      settings. Applying the KNN method in the resulting plane   
      gave 77% accuracy. However, these results are strongly
      biased (See Aeberhard's second ref. above, or email to
      stefan@coral.cs.jcu.edu.au). Results obtained by
      Aeberhard et al. are :
      RDA : 62.5%, KNN 53.1%, Opt. Disc. Plane 59.4%

      The data described 3 types of pathological lung cancers.
      The Authors give no information on the individual
      variables nor on where the data was originally used.

       -  In the original data 4 values for the fifth attribute were -1.
          These values have been changed to ? (unknown). (*)
       -  In the original data 1 value for the 39 attribute was 4.  This
          value has been changed to ? (unknown). (*)
   
      
5. Number of Instances: 32

6. Number of Attributes: 57 (1 class attribute, 56 predictive)

7. Attribute Information:

    attribute 1 is the class label.
   
    - All predictive attributes are nominal, taking on integer
      values 0-3

8. Missing Attribute Values: Attributes 5 and 39 (*)

9. Class Distribution:
    - 3 classes,
        1.)    9 observations
        2.)    13     "
        3.)    10     "