Since in auxiliary power unit (APU) fault prediction quick access recorder(QAR) data cannot meet the requirement of real-time or accuracy， an APU fault prediction based on real-time message data is proposed. First， the data collected by the message is preprocessed， and the message data of each flight is organized and put into one data set. Second， the data set is marked from the perspectives of parameter threshold， maintenance record and changes of APU sequence number. Third， aiming at the poor interpretation of the feature selection algorithm， we select parameters that can characterize APU performance through correlation analysis. Finally， a multi-parameter fault prediction model based on support vector machine(SVM) is established and optimized. It is proved that this model can improve the prediction accuracy and provide reference for APU maintenance strategy.