First a big thanks for the updated feature set of Encog, it means a lot to many of us that you continue to develop Encog furter! I think Encog is already today the best machine learning library, but since Christmas is around the corner I thought I'll let you know my wish list for next version already :-)
-There is a fairly new model out that has recently gained a lot of academic interest called Extreme Learning Machines.
This model has a similar result to SVM, but without the tedious training time. Training in ELM require only one iteration and one hidden layer so it can be 100s if not 1000s of time faster than SVM or MLFF for large datasets. Huang has developed both regression and classification version, there is a mathlab implementation at the site I referred to.
-The second feature I would like to see in Encog is component analysis methods. Working with financial data (and many other problems) we often see that data for different instruments are correlated, it would be good if Encog had the tools for doing this analysis. Although they might not be perfect Independent, Principal and Canonical component analysis would be of great value in order to understand the data and as input to the network. If you choose to implement ICA, I can suggest that you use the fastICA algorithm. There is a standard implementation that can probably be used as a basis http://research.ics.aalto.fi/ica/fastica/
-The last thing I would like to propose is methods for creating both ensemble networks (adaptive and bagging, suppose boosting would be useful as well) and deep learning machines.
Anyway, many thanks!