Getting different classification results while running the same SVM code.

Hi,

 I am working on SVM and I learnt that if some scaling is done for the attributes (features ) like getting their values
 from 0-500 to 0-5, the classification rate improves. I had put autoscale as false and done my own scaling as performance is poor in autoscale which is 50%
 I tried scaling it by 1000 and found that classification 
 improved from 50% to 75%. But when I am running the same code again and again, the classification rate is fluctuating between 70% to 80%. 

 Need your help to get rid of this.