As adaptive bitrate streaming services have spread, it has become more important for video streaming providers to control video quality and prevent viewing abandonments. However, since viewing abandonments are caused not only by quality degradations but also by a lack of users' interest in contents, it will first be necessary to clarify how quality and/or content affect viewing abandonments. To investigate this, we conducted an adaptive bitrate streaming experiment and developed a viewing-abandonment-reason-classification model that classifies abandonment reasons into quality or content. Using training data, we developed four models (logistic regression, classification tree, random forests, and support vector machine) where feature variables related to application quality, users' operation behaviors, and the attributes of viewed contents were used as explanatory variables. These four models were validated by using validation data. From the results, the support vector machine model was considered to be the best since it obtained relatively good validation results and did not appear to be over-trained.