Unknown aware k nearest neighbor classifier

Unknown awareness is very important for many applications such as face recognition. In a typical unknown aware classifier, an “unknown” label is assigned to strange test instances. This study proposes an unknown aware classifier known as UAkNN by extending the well-known kNN classifier. In UAkNN, unknown awareness is achieved by exploiting distances between instances of individual classes. These distances and their related statistics are used to confirm the kNN prediction or change it to “unknown”. Average accuracy of 85 percent based on the Iris dataset by using 5-fold cross validation has been achieved. Experimental results demonstrate that UAkNN is promising under various test situation.