Heuristic Classifier Performance Bounds in High Dimensional Settings
Paul M. Baggenstoss · 2002
This paper is concerned with probability density estimation in high-dimensional settings. Simplified geometric arguments and supporting examples point to a performance bound which limits algorithm performance to that of either (1) nearest-neighbor or (2) single-kernel PDF estimators. A method of monitoring PDF estimation performance as well as recommendations for neural net and classification algorithm practitioners is provided.