0

I am doing a project for classifying the presence of cars/bikes in an image.I have extracted the features from the images(data-set of cars and images not belonging to that of cars) and applied K-means clustering to get a uniform feature vector X for all images.Now upon applying Cross Validation( Folds=10) upon the data-set yields me an accuracy of ~61% when trained using SVM classifier( RBF Kernel).In this case,is reduction in the number of features a good option(before giving the feature obtained from image before giving it to clustering)?

Thanks and regards

logamadi
  • 27
  • 1
  • 5

1 Answers1

0

Yes you may examine the features on an individual basis to see if there is any relationship with the outcome of class labels.

But also make sure your parameters are optimal in your model. By implementing the parameter selection you are trading off between a larger margin and misclassified training points. Try tuning $C$ and $\gamma$ and observe the change of training accuracy.

Community
  • 1
lennon310
  • 2,582
  • 2
  • 21
  • 30