LAUSR

Machine learning methods for automatic classification problems using computational geometry are considered. Classes are defined by convex hulls of sets of points in a multidimensional feature space. The classification algorithms based on the evaluation of the proximity of a test point to the convex hulls of classes are examined. A new method for proximity evaluation based on linear programming is proposed. The corresponding nearest convex hull classifier is described. The results of experimental studies on real problems of medical diagnostics are presented. The comparison of the effectiveness of the proposed classifier with the classifiers of other types has shown a sufficiently high efficiency of the proposed method for proximity evaluation based on linear programming.