I’m recently working on time series data in my project, and I found that apart from traditional feature extract methods like PCA, there are deep learning methods such as auto encoders. I want to share in my group about time series data using these and make comparisons between them.
But you see they don’t predict results directly instead they make features.
So how can I compare their performance in features? Or to put it more frankly, how can we measure the quality of features we extract? As far as i googled there seems to be few articles about that.