From: thedevilsdisciple@gmail.com   
      
   On Dec 12, 3:13 pm, "Roberto"  wrote:   
   > Hi all,   
   > I'm develop a communicatino protocol, useful to wireless sensor networks,   
   > which uses SVM.   
   > In this work I use SVM in order to predict data in such a way to reduce the   
   > energy cost of each sensor in the network.   
   > So, I execute before a training phase on each sensor, while after I execute   
   > a prediction phase on each sensor. The predicted value is compared with the   
   > real value (about temperature, light, voltage, ecc...)   
   > However, I have seen that SVM Regression doesn't work very well.   
   > In fact, I my training set is about values taken from,  e.g., the period   
   > 08:00 AM to 12:00 AM and I would predict a value for 09:00 AM (i.e a time   
   > between 8 and 12 AM), regression works well.   
   > Instead, if I would like predict a value for 1:00PM (outside the previous   
   > interval of time) prediction doesn't work well.   
   > Is this normal?   
   > Does anybody have experience with this topic?   
   > Thanks in advance   
   >   
   > Roberto   
   >   
      
   Hi Roberto,   
      
   I guess the behavior of the system is non-stationary.   
      
   Also, any regression technique, be it mlr, neural networks etc., can   
   be expected to reliably perform only interpolation, and not   
   extrapolation as you are demanding.   
      
   Hope this helped.   
      
   Regards,   
      
   Denny Joseph   
      
   Reseach Engineer   
   Siemens Corporate Technology   
      
   [ comp.ai is moderated ... your article may take a while to appear. ]   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)