How can I effectively pre- process data?

I have an input-output dataset to be used in ANFIS modelling. I need to pre-process the data! I've already removed the outliers outside +/-3 standard deviation of the group mean! I've also employed the Moving Average Smoothing technique. The issue here is that the RMSE is still undesirably large! I need a more effective way to pre-process the data!

[Walter Roberson]

Is there good reason to believe that there is a meaningful amount of "signal" in the data? An actual pattern that is recognizable to be trained on?

[Olukayode Majekodunmi]

Yes! There is a significant amount of "Noise". It is noticeable from the plot. The system I'm trying to describe exhibit a high degree of non-linearity coupled with some errors incurred during measuring and recording! I discovered smoothing the data reduces the RMSE!

[Walter Roberson]

Perhaps you could give us an example plot or two? I would be interested in original data, then outlier-removed, then smoothed; and, of course, a line showing approximately what your ideal output would be.

http://www.mathworks.com/matlabcentral/answers/7924-where-can-i-upload-images-and-files-for-use-on-matlab-answers

[Olukayode Majekodunmi]

I'm sorry for the delayed response! The two inputs are to be modeled separately with each of the outputs (Sugeno Fuzzy). Here are the download links for the data plots! Figure 1: shows the original data http://www.sendspace.com/file/vsghku Figure 2: shows the data after the outliers have been removed http://www.sendspace.com/file/suiwet Figure 3: plot after applying the moving average smoothing technique http://www.sendspace.com/file/1svdkd Figure 4: after scaling the smoothed dataset http://www.sendspace.com/file/1gs4yi