🎉 🎉. RAPIDMINER 9.8 IS OUT!!! 🎉 🎉
RapidMiner 9.8 continues to innovate in data science collaboration, connectivity and governance
feedforward backpropagation learner
For traffic data prevision and network capacity planning we are looking for a good backpropagation feedforward learner in rapidminer. We have been trying to use the W-Multi-layer Perceptron, but the results are not satisfying (error rate too high -> poor prediction results). In addition, it is not very clear in Rapid-i how to set and tune the parameters of this network. Qustion: which backpropation feedforward do you propose and is there any information about how to get good results out of it..? Our training- and testdatasets are ok. Thanks a lot for helping us one step further!
Answer by Ingo Mierswa:
we provide two neural network learners: the one of Weka you used and the operator NeuralNet which is based on the Joone neural network library. You could checkout the NeuralNet operator, too. Personally, I found it easier to define the parameters for the latter.
You could also try other (loosely) related learning schemes. I would almost always prefer SVM over neural networks and noticed that SVM often outperform them so I would probably try some of the SVMs, too.
For prediction settings related to series data you could basically use any regression learner in combination with the windowing operators provided by RapidMiner. Since there is no silver bullet, you just have to try which combinations turns out to work best.