CBR for road conditions on Norwegian alpine roads

Abstract

This study s objective was to investigate various improvements to and the generali-zation of an existing CBR system for road conditions. Results indicate a slight per-formance improvement from using k-nearest neighbors together with a previously designed decision dependent similarity measure. The generalized version of the CBR system appears to perform well, and tests show potential for using the system on different alpine roads. An overweight of cases where the road is open as op-posed to closed, seems to introduce a bias towards predicting that the road should be open. This is also a challenge for the methods for tuning and maintenance tested in the study. Further work remains with respect to creating a full blown CBR system.