well what do you want to get in the end? I mean, what do you want to achieve?
With MLP Network you can predict a class (nominal) or numeric value. This is called supervised learning due to training data for the learn process. You learn a rule or a function.
Okay you want to detect fraud. Is that a classification procesdure? If you have a training set (that means you have datasets where you already have the answer if it was fraud or not) you can use a supervised learning method to predict the results for datasets where u don't have the answers.
The fuzzy clustering is unsupervied learning (to detect patterns (here: clusters)) where u don't have a dataset with results (no training data).
I have no idea why you want to combine these methods and what for?
There is a technique called hybrid learning (used in RBF networks for example). There you first cluster the data and then predict the results based on a distance meassure between cluster centres and input.
Just tell us what data you have got and what you want to get. I'll help you...
Q1) well what do you want to get in the end? I mean, what do you want to achieve?
the answer to this question is that i want an ROC curve that detects true positives at a higher rate than false positives.
Q2)Is that a classification procesdure?
ans- yes it is a classification process. i have to divide my data into two types- fraudulent and non fraudulent. and i do have a training set.
Q3) I have no idea why you want to combine these methods and what for?
ans) my research topic is telecommunnication fraud detection using neuro fuzzy logic... as the research has already been done on this topic using neural networks only so i want to add fuzzy logic to it..
now one more question that i want to ask is how does rbf learning technique work?? actually i m not getting how to represent my output using this node.... pls help...
hm thats kinda strange. I means you want to use fuzzy logic because thats your research tpoic, but you don't really know what it is? Maybe you should first make things clear with your professor since I think it is not the right way just to try the nodes in knime and don't know how it really workes (behind the scenes) and it is not a scientific way.
When u google for "fuzzy logic neuronal networks" there are a lot of articles about it.
check this out:
http://portal.acm.org/citation.cfm?id=126187
sounds like what you are looking for.
Fuzzy logic is not only used by clustering, maybe thats the point here.
they predict fuel lubricity by using a k means clustering and then feed the rbf network with these centre values to predict.
In KNIME you can use the RBF network by getting the weka plugin. There is a kmeans clustering already implemented in this node (if you want to use other things like fuzzy clustering you may have to change the implementation). Feed the node RBFnetwork (learner) with the training data and select the class column. Set your paramters (std. dev and number of clusters). use the weka predictor and feed it with your test data.
Here u find a nice java applet which lets you play around with number of clusters and std dev.
actually i tried to figure out the working of nodes in knime... bt i m unable to locate the files that can explain what all is happening at backend...:(