A Study of Complications in Real - world

This paper is a case study of a long-term project studying the application of machine learning to the problem of dispatching technicians to x faults in the local loop of a telephone network. The bottom line of the study is that simple machine learning techniques can increase the eeectiveness of local-loop dispatch. However, presenting a convincing argument to that eeect has been far from simple. The case study shows how almost all of the complexity in this application of machine learning is related to issues that are peripheral in the machine learning literature. In particular, we had to gather data from multiple sources, use domain knowledge to clean up data, and deal with the fact that an evaluation based on classiication accuracy is relatively meaningless without an analysis of cost eeectiveness. Our view is that application studies should be helpful in guiding future research. Therefore we conclude by outlining useful directions suggested by our experience on this long-term project.