Network traffic is evolving by the evolution of networking technologies. Organizations and governments are getting suspicions from the type of traffic in their infrastructures because of the sensitivity of the data stored in each. Network traffic classification was one of the suggested solutions to protect such type of data. With the machine learning ability to handle huge amount of data to analyze and monitor, it was widely used as a method to increase Confidentiality, Privacy and integrity. In this paper we discuss the K-Nearest Neighbor machine learning algorithm in classifying network flows to detect malicious activity in an organization infrastructure. KNN perform a training phase to let the system detect the dataset types and in the testing phase it can classify unknown flows relying on the training set performed. However, since knn is a lazy classifier, adding weight to the training phase increase classifier accuracy such as sigmoid function. Our method focuses on enhancing the sigmoid weighting criteria to increase the WKS model accuracy. Boosting the sigmoid weight depending on the correct comparison in the training set improved model accuracy by analyzing several evaluation parameters from the weka application. Result were promising and proves the enhanced model accuracy over the standard WKS model by several evaluation parameters. However, the time to perform the classification increased as well