I think I can find Alternative Splicing Sequences and data from ProSplicer,
to prove my model,
then train my model.

today NetGene2 not work, so I try to find others.
http://l25.itba.mi.cnr.it/~webgene/wwwgene.html
http://bioinformatics.iastate.edu/cgi-bin/sp.cgi

maybe try if "exonic splicing enhancers predictor" can help improve ours or not.
ref: http://nar.oupjournals.org/cgi/reprint/31/13/3568.pdf

this site do the same thing as me, connect to other sites to obtain interstat values.
ref: http://www.inra.fr/bia/T/schiex/Export/LNCS-EuGene.pdf

this software is a A Generic Framework for the Integration of Gene-Prediction Data
ref: http://www.genome.org/cgi/reprint/12/9/1418.pdf