[StevenB]

Instead of stopping as soon as you detect a screw, you could wait for two (or three, or whatever) consecutive "screw present" readings. This would eliminate spurious stops from a single noisy reading. If necessary, you could stop the motor, collect a few readings, and then make a decision about whether there was a screw or just noise.

Because you know approximately how long the motor should run between screws, you could also filter out detections that come far too soon. However, this only reduces the problem; it doesn't really eliminate it.