I solved select cancer gene feature completely and publisher the book from Springer at the end of this Jan. And I finish to write cancer gene diagnosis book already.
I plan to publish it within this year. See my RG HP.
I think you may apply......but it characteristics totally depends on the nature of the gene data which you are using......You may follow these below links for your help......
I solved select cancer gene feature completely and publisher the book from Springer at the end of this Jan. And I finish to write cancer gene diagnosis book already.
I plan to publish it within this year. See my RG HP.
First you calculate correlation between different features and and check for both negative as well as a positive correlation coefficients. If feature shows a higher correlation with other features, then you will check the performance of these features individually and in a combined way and select a feature or group of features, which soever perform better.
If you want to know more detail about implementation please mail me.
In any case you can apply CFS (I guess you use Weka tool). In case both the features and the targets are continuous the algorithm uses PEarson correlation, if the target is classification, it uses Symmetrical Uncertainty (which is based on mutual information) after discretization of continuous features.