Internet technology continues to revolutionize real-time access to online videos, television shows, and sporting events. The authors of this paper echo the conflict in expectations between service providers and Internet video consumers. Content and player designers, providers, and video network distribution centers are concerned about the performance of video delivery mechanisms, based on measures such as startup delay, buffering, bit rates, and frame depiction rates. However, depending on their interests and video viewing patterns, Internet video customers may be jubilant or fuming over high bit-rate switching and the content distribution network’s performance. Even so, how should the interests of online video customers and video content designers and distributors be reconciled for a win-win situation? What quantitative and qualitative models should be used to forecast customer interest and actual access to Internet videos? What metrics should be made available to video customers, for use in comparing the service delivery of alternative video content distribution centers?

The authors present perceptive ideas on the design of a model for forecasting playtime and frequency of access to videos by consumers, based on measures of the quality of access, such as buffering and bit rates of the video content distribution networks. The predictive model they propose is similar to the well-known neural network model with forward input and back-propagation learning [1]. The authors use simple regression analysis and binary decision tree theory to investigate the relationships and predictive aspects of user playtime and access to online videos, based on the video access quality metrics offered by content distribution networks.

The authors did not find any statistically compelling variance in online video playtime and access attributable to the quality of online video technology. This is not surprising given that simple regression and the decision tree model are inadequate for the analysis of nonlinear data. The usage data ought to be normalized, recoded, or transformed. Statistical models capable of logistic regression and discriminant analysis [2], which support the sensitivity analysis of “what if” questions, could then be used to analyze user engagement. Clearly, the authors advocate for and offer new insights into constructing unified video quality metrics for both users and Internet video distributors.