Dear all;
I wrote some codes for fault detection in wind turbine with SVM classifiction. I use matlab 2014b.
I totally have 24002*6 data in a normal class and a fault class. I performed PCA to this data I reduced features to one ,then I used 9601 data in a normal class and in addition 9601 data in a fault class for training , 2400 data in a normal class in addition 2400 data in a fault class for test in SVM classification.
My problem is that when I perform PCA for dimensionality reduction my codes don't run and it takes more than 10 minutes!
Can anyone helps me to solve this problem?
Your help is appreciated
Hi Faeze,
Can you explain a bit more about your implementation. Is this 10 minutes the running time of PCA+SVM or only PCA?
BTW, if the data is separable in the one-dimesional feature space, wouldn't it be more efficient to replace SVM with faster methods such as LDA?
Well.. try to debug it see what are the cause for slowing down. Maybe you configure wrongly. If there is nothing wrong about algorithm configuration or hyparameter tuning. Try to not only reduce the dimensionality but also the amount of training instances. If you can use small portion of training data to learn, DO NOT USE the full training data as the more the data you use, the higher chance outliers ore redundant data may effect the quality of the model.
Well ! U Can Parallleize the Codes using CUDA / Open CVprogramming plateforms
Prof K S Sridharan
Article SVM Multi-Classification of Induction Machine's bearings def...
- Badges
- Science topic