ANOVA-SVM for Selecting Subset Features in Encrypted Internet Traffic Classification

Abstract

Encryption technique is widely used in the internet network for protecting user privacy, maintaining the confidentiality of the data, avoiding firewall detection, and administrating the system. To prevent encryption techniques in malicious activities such as encrypting data that contains malware or viruses, illegal transactions like selling drugs, illegal weapons and fake documents, a company or institution uses encrypted internet traffic classification to analyze and identify the activity. A challenging problem in encrypted internet traffic classification is the massive amount of data in the dataset and the existence of many irrelevant features. In this research, we propose a technique by integrating the ANOVA algorithm with the wrapper method from LinearSVC in the SVM method to overcome this problem. The ANOVA algorithm is used to analyze the data's variance, and LinearSVC to calculate the relationship between each data to its decision boundary. A new technique is proposed by calculating the mean value of the distances to remove features, which are relatively far from the decision boundary. This technique is taken to isolate features from unused ones to be used for the next process. The experimental result shows that this proposed method can compete with the existing research method and reduce system detection time. In this case, we take some research as the baseline, including that with one-dimensional convolution neural networks, over-sampling and undersampling combination, inline and adaptive application, and FlowPic.

Machine learning corroborates subjective ratings of walking and balance difficulty in multiple sclerosis

“…We evaluated the feature importance through three different types of models, intended to check which features are consistently important between models. The pipeline ANOVA-SVM (Megantara and Ahmad, 2021 ) calculates the average F-score for the selected features. Linear models calculate the coefficient of each feature to determine feature importance.…”

Optimizing Feature Selection Method in Intrusion Detection System Using Thresholding

“…ANOVA-f test. Analysis of variance, or ANOVA, is a method used to determine any statistically significant differences among the means of two or more groups into the target variable [33]. The high value of F-statistic and low p-value suggests that there are significant differences of the groups, suggesting that the features are relevant to the targeted variable [34].…”