This paper proposes an algorithm called Predictor. This algorithm uses an automaton per matched episode rule with general form. With the aim of finding the latest minimal and non-overlapping occurrence of all antecedents, Predictor simultaneously tracks the state transition of each automaton by a single scanning of data stream, which can not only map the boundless streaming data into the finite state space but also avoid over-matching episode rules. In addition, the results of Predictor contain the occurring intervals and occurring probabilities of future episodes. Theoretical analysis and experimental evaluation demonstrate Predictor has higher prediction efficiency and prediction precision.