XPost: comp.ai.neural-nets, sci.image.processing, sci.math.num-analysis   
   From: milind.a.joshi@gmail.com   
      
   Hi Peter,   
      
   It does look like you've done some pretty good analysis. PCA might work   
   well, and so would a whole list of other approaches. Have you tried   
   boosted variants of simple algorithms?   
      
   Simple things go a long way: Ensure that you don't train on all your   
   data, for example, and if possible, have some data that even you   
   haven't seen before available.   
      
   Another thing you could do is use some ready-made tools to do the basic   
   tests, before you go on to your own implementation, this would save you   
   a lot of time and speculation.   
      
   In the end if your work is intended for practical applications, you   
   will probably have to make tweaks and adustments that go beyond pure   
   algorithms, and will match your domain and data really well, but might   
   have no application outside of your problem domain.   
      
   Also, you will probably tweak your approach, uses mixture models, and   
   so on, over a period of time.   
      
   Good luck!   
      
   Regards,   
   Milind Joshi   
   IDEA TECHNOSOFT INC.   
   http://www.ideatechnosoft.com   
      
   [ comp.ai is moderated.  To submit, just post and be patient, or if ]   
   [ that fails mail your article to , and ]   
   [ ask your news administrator to fix the problems with your system. ]   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)