Generally, microarray data sets are quite large in size. So, it will be difficult and time consuming as well to apply Chi square/Gain ratio algorithms. So, I think that you may calculate the variance and standard deviation of your data sets and organize them in either ascending or descending order.After doing that, if you apply feature selection algorithm....you will get more good result in terms of accuracy and also the size of the database will be reduced if you omit those records whose variances are less.

Thanks Dr Sanjay 

Actually,  I am going to form ensemble feature selection algorithm from different metric .Thereby, I calculate the average of scores for all features , so, need the score for all features from IG and gain ratio..

Hi Osama,

Normalizing the data is not necessary, as this depends on your task, from the one hand, and data type, from the other hand. Nevertheless, normalizing the data will not solve your problem. 

However, "ChiSquaredAttributeEval" didn't work in your case because your dataset may have, e.g., numeric class. In such case, even if you're going to normalize the data, ChiSquaredAttributeEval will not work. To solve this matter, you need to convert the class type into either binary class or nominal class or discretize this class (if you want to use "ChiSquaredAttributeEval" in WEKA correctly). In addition, you have to use "Ranker" method in conjunction with "ChiSquaredAttributeEval" in order to perform complete attribute selection with WEKA.

HTH. 

Samer