|
I used both KNN and SVM for classifing hand shapes using images from the Cambridge Hand Gesture Dataset. the results obtained by the KNN method were better than the results from the SVM classifier. I was asked to give an explaination for this. I only figured that out from the experimantal results and I don't know why it gave those results. |
Did you do cross validation
Did you use any kernel for SVM ?? Non - linear separation could be one cause for this. Nearest neighbour is more robust if you have few classes and more data with less variation.
Have a look here anout benefits of kNN http://fastml.com/the-secret-of-the-big-guys/