An Automatic Error Identification Method in Call Control Protocol Using Levenshtein Distance

As the number of IP-based telephony-like voice over IP (VoIP) users increases, interoperability of devices will become much more important to achieve smooth communications between arbitrary pairs of devices. To this end, detection and analysis of communication failures are key issues for validating interoperability. However, current failure (error) detections must be manually confirmed, which requires significant effort. Therefore, an automatic error-identification method is necessary. A straightforward approach can be considered that involves using a learning-based algorithm (e.g., k-nearest-neighbor-based classification) to identify errors; however, it is difficult to collect useful training data to achieve high identification accuracy. Since the number of conditions that we can prepare for in advance to collect training data is quite a bit less than that in a real environment, many errors may be misidentified when such cases are not included in the training data. To solve this problem, we propose a method which considers the characteristics of message sequences in a VoIP session control protocol (i.e., Session Initiation Protocol) by applying a similar approach of editing the distance between strings. Experiments using real environment data show that our method can improve the accuracy of identifying errors that are not included in the training data and enable administrators to collect new training data for unknown errors.