Cody M. Lewis


Game Balancing Machine Learning Used On Security

Without Balancing

With Balancing

Explanation

This is an abstract simulation of a store surveilance system, the blue squares represent non-malicious shoppers, the red squares represent malicious shoppers, the squares surrounding them are the systems classification of them.

This demonstrates the difference made when using balancing techniques in evolutionary algorithms similar to those suggested for video games.

The balancing classfier is made to be less sensitive. Once it identifies a person that it believes to be suspicious, it will also mark a random node as suspicious in order to bring balance.

The balancing does not seem to improve accuracy, but brings a guilty by association property to the classifier, which may be useful in situations such as surveilance systems for areas with high gang activity.