What is the minimum sample of SVM?

Joanneyu

Hi, 
I use Auto Model to conduct SVM regression. I wonder apart from 100, which is the minimum sample needed for Rapidminer, is there any way to justify the ideal number of SVM to provide reliable results? Thank you!

hughesfleming68

Ask five different people and you will get five different answers. My point of view is that too much data is just as unhelpful as too little. Older data may not be relevant to current conditions. If your data is relatively stable then usually more data is usually better. The only way to know with your data is to vary the look back length with the risk being over fitting. If your accuracy improves a lot with with 101 samples then it could be just an anomaly in the data. If you look at examples of sliding window validation, you will often find quite small windows like 10. These decisions are all data dependent.