This paper discusses the use of support vector machines (SVMs) to detect and predict the health of multivariate systems based on training data representative of healthy operating conditions. This paper also investigates a novel approach to SV classification and regression through the use of a principal component projection pursuit. Statistical indexes extracted from the reduced input space are used in a time series fashion for SV regression to predict the system health. The approach benefits from the reduced input space and from the small number of support vectors used to construct the classifier and predictor models, making it faster and robust. It is also immune to probabilistic assumptions and to the need for explicit models that describe the system behavior. A case study illustrates the use of support vector classification and regression. Case study results show that SVC correctly classified test points and minimized the number of false alarms and SVR correctly predicted function values for a predefined sinusoidal function. Together with excellent generalization ability, the proposed algorithm can be used in real time, making it a strong candidate for on-board, autonomous, system health monitoring, management and prediction.