The infinite number of websites in the internet world can be classified into a number of categories depending on the purpose for classification.In educational institutions, it is required that the internet facility be used for academic purposes only. This can happen through accurate classification of network traffic into two classes, educational websites and non-educational websites. In educational institutes, for the optimum use of network resources the use of non-educational websites should be banned. In our research work, we explore the use of ML (machine learning) algorithms for classification of internet traffic into two classes. We discuss the basis for differentiating the traffic into these two classes, and investigate the impact of flexibility in making such differentiation on the performance of selected five ML algorithms. We also investigate the impact of class error in data set and propose a generalized feature subset which is robust against all these impacts. Results show that Bayes Net algorithm performs consistently for intended classification of internet traffic in terms of classification accuracy, training time of classifiers, recall and precision.