PCA huge parts of missing data filling

Question

I’m performing PCA on different time series’ and then using K Means clustering to try and group together common factors.

The issue I’m facing is that some of the factors come in and out of the time series. For example I may have 12 years in total of data points, some factors may exist for the entire 12 years but some may dip in and out (active for the first two years, inactive for three years, active for the rest for example).

I can use iterative PCA to fill in some of the usual gaps in data but I am unsure of what to use for the large chunks (years) of data gaps. For example if a factor only exists for the final three years should I even incorporate it into the analysis?

Answer 1

Principal component analysis (PCA) is not designed for time series data. It would be better to switch to singular spectrum analysis (SSA) which is designed for time series.

There are gap-filling versions of SSA for missing data.