Early Transition Detection - A Dynamic Extension to Common Classification Methods

Abstract An extension for classification methods in order to process time-dependent data is introduced. It is based on the detection of transitions from one steady state to another one by examination of the time derivatives of classification vectors. The method is called Early Transition Detection (ETD). It is shown that it can be used in conjunction with a number of common classification methods like SIMCA or Artificial Neural Nets and it is successfully tested on simulated and on real data.