Surge is a phenomenon of large oscillations of pressure and flow that occurs in dynamic compressors when the compressor is operated at too low a flow rate for a given discharge pressure. In order to prevent damage to the compressor, surge must be avoided. A typical surge prevention measure for compressor operation is so-called surge avoidance control, which normally relies on a map of the operating conditions for surge occurrence, i.e. surge map or surge line. The surge map currently used in industry practice is insufficient since surge occurrence is affected by many more process variables, especially ambient air conditions. If large uncertainty exists in surge conditions, the compressor has to be operated conservatively far away from the surge line, which limits the dynamic range and usually sacrifices the efficiency. In order to obtain more accurate surge map adaptively, a data-driven surge map modeling approach has been developed using support vector machine (SVM) approach based on surge test data. Surge map is obtained as the classification surface between surge and not-surge data. Principal components analysis (PCA) is used to identify the variables that contribute most to surge. The method of asymmetric support vector machines (ASVM) is developed to reduce the possibilities of missed surge prediction. The developed methodology was verified with the actual testing data on the centrifugal air compressor in Toyota Motors.