Monitoring Unhealthy Hosts in Application ELB Target Group in Cloudwatch

We have a cloudwatch monitor setup where an alert is fired if the number of unhealthy hosts is < 2 for 120 seconds. Over the past week or so, an alert has been fired and when I go into cloudwatch I see that the value for Unhealthy hosts is 1.83 or 1.66, I thought it would have been an on or off thing, and go to 1, not some random decimal in between 2 and 1. Has anyone see this behavior before?