Supervised Learning to Predict Human Driver Merging Behavior

This paper uses the supervised learning techniques of linear regression and support vector machines in an attempt to predict the merging behavior of drivers on U.S. Highway 101 based on current traffic patterns. Using highway data from the Federal Highway Administration, the paper approaches the problem from numerous angles, eventually concluding that using a twostep approach achieves the best result. The first step is to cluster the drivers based on driver history, and train separate models for each of the clusters. This results in the best results in every case.