XGBoost - Array of (sparse) time series as one input to classifier

Question

I have online activity data, which for each user consists of N (like e.g. N=410) very sparse time series, spanning some M instances (M can be tens or can be thousands, not so im