Classification of bandwidth-heavy Internet traffic is important for network administrators to throttle network of heavy bandwidth traffic applications. Statistical methods have been previously proposed as promising method to identify Internet traffic based on their statistical features. The selection of statistical features still plays an important role in accurate and timely classification although most feature selection algorithms consider the correlation between features. In this work, we propose a technique based on features characters and Principal Components Analysis (PCA) feature selection algorithms for online Peer-to-Peer (P2P) traffic detection. Using Naïve Bayes and J48 machine learning techniques for available traces from University of Brescia and University of Cambridge, experimental results show that the proposed method is able to achieve up to 99.5% accuracy for 0.007 second testing time. These results are superior to other existing approaches in term of accuracy and testing time.