In a practical optical fiber perimeter security system, not only the discrimination of multiple events but also the comprehensive probability evaluation of these events is required. Therefore, this paper proposes a recognition scheme combining autoregressive moving average (ARMA) modeling with Sigmoid probability fitting. In event discrimination, both the ARMA coefficients and the zero-crossing rate of an optical fiber vibration signal are incorporated into a feature vector, which is then fed into a support vector machine (SVM) to recognize six types of common intrusion events: climbing, knocking, waggling, cutting, kicking, and crashing. In comprehensive probability evaluation, the SVM training pattern outputs are used to fit the parameters of a Sigmoid function. Then, the SVM outputs of the test patterns are substituted into this fitted Sigmoid model to yield the expected result. Field experiments reveal that the average recognition rate of six intrusion events by the proposed scheme reaches 87.14%. Moreover, the occurrence probabilities of all intrusion events can be provided as references, thereby presenting vast potential for future applications.