I have been building models with weka nodes in KNIME. The workflow starts from reading sdf files to weka predictions. I dont understand why, i get 100% prediction with all weka methods and with different attributes, test set, 5 folds, 10 folds and 100 folds. I knew that there is some problem in data but i could not find it. Descriptors were calculated from MOE software and then normalized with knime node (i tried two methods, one is min-max and another one Z-score), subsequently models were developed with weka nodes.
These normalization methods are giving 100% prediction!
i have tried same data and normalized in weka package and models gives 30% prediction (this was i expected, becuase data is not good).
Do you know, what is happining in normalization methods in KNIME ? because it is more important to my project (i have developed more than 1000 models from last two months).