Constructing Reliable Classifiers for Road Side Assistance

This paper considers the problem of constructing classifiers for road side assistance capable of providing reliability values for classifications of individual instances. In this context we analyze the existing approaches to reliable classification based on the typicalness framework [14, 25]. As a result we propose an approach that allows the framework to be applied to any type of classifiers so that the classification-reliability values can be computed for each class. The experiments show that the approach outperforms the existing approaches to reliable classification for road side assistance.