hi,

after trying svm and NN for identifying patterns, but failing in real and live trades badly, I was thinking of using fuzzy system. does any one had the chance to evaluates their performance in live trading..
thanks..

There is a thread in the elite section on this topic. But a quick thought would be the net over-fit your data. If you training set reported good results and validation set didn't, this is usually the case.

yes I think over fit is different. however after experimenting with some parameters like learning rate, performance etc. the results are not bad.

There are 02 types for fuzzy systems.

a. Mamdani Method
b. Sugeno Method

how the inputs translates to output is determined by membership function (mf).
so far I have seen Mamdani type FIS structures to be giving better results in trading scenarios.

as far , For Sugeno I am in the middle of experimenting.

based on my personal learning curve and observations,

you need to fit the MFs very accurately according to data. for ins and outs.

however plus points are you can adjust the weights for rules.

If every thing done carefully to build the system. it can provide very good results.

It has fixed weights, linear only MFS are possible. computationally efficient, and results are good as well.

The neuro fuzzy systems use back propagation to learn, and these systems may prove better than all other AI structures, and during learning If system training error is decreasing, and system checking error starts increasing, it is a clear sign of overfitting...