The aim of the challenge is to predict crime hotspots which for the most part don't vary much I'd think? The challenge then becomes identifying cluster density as a function of time which might mean using past data makes sense.