Critical Zone Recognition: Classification vs. regression

The article describes the Classification and Regression procedures, developed and successfully used for Critical Zone Recognition. One of the main tasks of the Prognostics and Health Management is the Failure Prognostics, specifically to provide predictive information regarding Remaining Useful Life (RUL) of device using prognostic systems. But sometimes it is necessary to get inflexible answer for closed type question: Is current device within critical zone or not? In other words, is RUL of device less than pre-defined Critical Value or not? To solve this problem, two approaches may be considered: · Regression Approach: to predict RUL value and compare results with critical value · Classification Approach: to recognize directly entering the critical zone In general, Classification Approach is more preferred for recognition tasks, but some aspects of the approach prevent to get an evident answer. Two models, based on modifications of the SVM method - SVC (Support Vector Classification) and SVR (Support Vector Regression) are proposed for consideration. Suggested methodology and algorithms were verified on the NASA Aircraft Engine database (http://ti.arc.nasa.gov/tech/dash/pcoe/prognostic-data-repository/). Numerical examples, based on this database, have been also considered.