The enormous growth of streaming services in the last decade leads to the emergence of the Quality of Experience (QoE) metric, which aims to improve and optimize the delivery of video streaming service, thus strengthening the loyalty of end-users to the provided services. Yet, predicting QoE of a multimedia stream is a challenging task because it is dependent on several different influencing factors. Moreover, it should handle dynamic environments with large-scale data. Machine learning methods offer a method for quantifying the intricate connections between various influencing factors and QoE. Thus, in this paper, a new online QoE prediction method is proposed, namely, Incremental Stacked Support Vector Machine (ISSVM). The proposed approach uses a developed stacked generalization technique to increase the global accuracy and minimize the execution time, by combining predictions of several parallel Multi-class Incremental SVM (ISVM) learners trained with different types of sub-features. Then another ISVM model is used as a meta-classifier instead of a simple linear regression model in order to build a robust fully incremental model. In fact, using the ISVM model as weak classifiers aims to handle non-stationary and very huge volumes of data in real-time contexts. The findings show that the suggested model is more effective over the rest of the state-of-the-art methods.