A method and system are described for end-user transaction recognition based
on
server data such as sequences of remote procedure calls (RPCs). The method may
comprise machine-learning techniques for pattern recognition such as Bayesian classification,
feature extraction mechanisms, and a dynamic-programming approach to segmentation
of RPC sequences. The method preferably combines information-theoretic and machine-learning
approaches. The system preferably includes a learning engine and an operation engine.
A learning engine may comprise a data preparation subsystem (feature extraction)
and a Bayes Net learning subsystem (model construction). The operation engine may
comprise transaction segmentation and transaction classification subsystems.
