While CleverLoop is capable of learning about the environment your cameras are viewing and how you react to alerts and movements, as well as having fine tuning options to reduce false alarms, there are some things that you some things that you can't teach it to do, and there are others that are possible but you should avoid.

You can't tell your CleverLoop system that a person should be a 'movement' rather than an 'alert'. 

Your CleverLoop system can tell when it is seeing a person, but it has no way of telling if a person is a stranger/intruder, or say a member of your family for example. If you want to stop getting alerts about people who are supposed to be in the areas that your cameras are covering, then you need to use the Geofencing and Scheduling functions in the Auto-Arm Settings page. Geofencing triggers CleverLoop to automatically disarm when those family members and friends who are logged into the app enter your home. 

A camera becomes less sensitive to important movements

Repeatedly telling your CleverLoop system that significant events are 'movements' rather than 'alerts' can lead to the system becoming less sensitive to things that you want to be alerted about. If this happens, then you need to use the 'Forget all learned patterns' option in the settings for that particular camera, and start the teaching process again.

Your cameras are having problems identifying partial views of people

The camera needs to see as much of a persons body as possible for the algorithm to recognise them accurately as an 'alert'. If a camera is likely to only see part of a person, say through a window, then you can tell the app this for each camera in the 'Fine-tune Smart Detection' section of the cameras settings. This setting is only available if you select the option of the camera facing inside.

Try changing the placement of cameras that are having trouble so that they have better views of the areas you are trying to protect. See this article on camera placement.