Kernel PCA for novelty detection

Novelty detection indexes are used in order to identify anomaly in the observation of a phenomenon. We describe the basic idea of kernel principal component analysis, a method which enlightens the existence of a novelty in a measured value comparing it with the one predicted by a model calibrated on training data. Differently from linear PCA, kernel PCA projects the data into an infinite-dimensional space in which novelty detection has usually a better performance.