The concept, structures, and algorithms of principal feature classific
ation (PFC) are presented in this paper. PFC is intended to solve comp
lex classification problems with large data sets. A PFC network is des
igned by sequentially finding principal features and removing training
data which has already been correctly classified. PFC combines advant
ages of statistical pattern recognition, decision trees, and artificia
l neural networks (ANN's) and provides fast learning with good perform
ance and a simple network structure. For the real-world applications o
f this paper, PFC provides better performance than conventional statis
tical pattern recognition, avoids the long training times of backpropa
gation and other gradient-decent algorithms for ANN's, and provides a
low-complexity structure for realization.