A novel feature line segment approach for pattern classification

In this paper, a novel pattern classification approach is proposed called shortest feature line segment (SFLS). It retains the ideas and advantages of nearest feature line (NFL) and it can suppress the drawbacks of NFL, i.e., the extrapolation inaccuracy, interpolation inaccuracy and high computational complexity. SFLS uses length of the feature line segment satisfying given geometric relation constraints, instead of the perpendicular distance from query point to feature line in NFL. SFLS has clear geometric-theory foundation and its implementation is relatively simple. In experiments based on artificial datasets and real-world datasets, comparisons between SFLS and other classification methods are provided, including nearest neighbor (NN), k-NN, NFL and some refined NFL methods. Experimental results show that SFLS is a simple yet effective classification approach.