A new probabilistic classifier based on decomposable models with application to internet traffic. (May 2018)

• Record Type: Journal Article
• Authors: Ghofrani, Fatemeh; Keshavarz-Haddad, Alireza; Jamshidi, Ali
• Is Part Of: Pattern recognition. Volume 77(2018:May)
• Page Start: 1
• Page End: 11
• Publication Date: 2018-05
• Subjects: Decomposable models -- Over-fitting -- TAN based on Chow–Liu algorithm -- Averaged TAN -- t-Cherry algorithm -- Internet traffic classification; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2017.12.009
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: In this paper, we proposed a new algorithm called LDMLCS to build a family of decomposable models that are appropriate for classification process. LDMLCS can generate numerous decomposable models, including simple models such as Tree Augmented Naive Bayes (TAN) and complex models with large joint marginal distribution, and some models that could be used for classification on a given dataset. This algorithm can address the problem of over-fitting and also capture the interaction between different features effectively and consequently, the obtained model is easily interpretable. Abstract: Probabilistic models are one of the common approaches for classification. One way to create these models is to form a decomposable model by selecting a set of marginal distributions. Although decomposable models are attractive by having some desirable properties, they would face the over-fitting issue in model-based classification approaches. Considering this issue, we propose a new method for selecting a set of marginal distributions and creating a proper decomposable model while controlling the complexity. The obtained model will be able to capture the interdependencies among different features and can be used for classification. The proposed method is compared with three existing methods namely TAN and Averaged TAN classifiers and t-Cherry algorithm by focusing on internet traffic data. Experimental results show that our obtained model can effectively extract dependencies among features, and hence, its performance as a classifier is superior compared to other three methods.