I have daily data of length 15k around. I need to use LS-SVM for forecasting. If I give approximately 10K data for training the systems runs out of memory. I checked that LS-SVM becomes really slow if I want to train it using many data points. Is there any solution that I can save my time. Is LS-SVM really performs well for data points less than 3000 (as it was advised to use data length less than 3k for computation).