Data Mining Approach To StudyQuality Of Voice Over IP Applications

Abstract

Patterns focus on “interesting” structures in data behaviour, which can be abstracted and used for efficient understanding and analysis. This paper presents a data mining technology aimed to assess characteristics of network connections for Voice over IP (VoIP) communication based on analysis of structures of Quality of Service (QoS) parameter time series data (end-to-end delay and packet loss). The data mining approach for QoS parameter data is designed and developed in the framework of EU IST INTERMON project [1]. It is used for spatio-temporal analysis in the large scale inter-domain Internet. We propose automated techniques to study network QoS parameter patterns for VoIP communication, such as: “Outliers” defining QoS parameter values exceeding “thresholds” corresponding to the rating factor R of the ITU-T E-model. End-to-end delay structures, used for appropriate choice of playback delay adjustment algorithms. Based on patterns, the network administrator can automatically evaluate the ability of network connections for VoIP communication, and select appropriate parameters for playback algorithms. The discussed data mining facilities for the QoS parameter study for VoIP applications are shown using real measurement data for inter-domain Internet connections collected in the INTERMON data base.