I would methodically break these down and review each type in context. Then look at the heavy hitters. For example if your threshold is only 250ms for x alert and you have a lot in the 300-400ms range, then you can either ignore those for now and/or raise that threshold.

Once you've identified the worse offenders look through the PDN for each alert type for what they mean, and recommendations on how to diagnose them.