ICTACT Journal on Soft Computing

T. Subbulakshmi ¹, R. Regin Raja ¹

Affiliations
1 School of Computing Science and Engineering, VIT University Chennai Campus, IN

Abstract

Sentiment denotes a person's opinion or feeling towards a subject that they are discussing about in that conversation. This has been one of the most researched and industrially promising fields in natural language processing. There are several methods employed for performing sentiment analytics. Since this classification problem involves natural language processing, every solution has its own advantages and disadvantages. Hence mostly, a combination of these methods provides better results. Various such ensemble approaches exist. The objective of this work is to design a better ensemble approach that uses a complex voting method, where classifiers are given rights not only to vote in favour of classes but also against them. This in turn will give chances to the algorithms that are weaker in classifying a sentence toward a particular class but better at rejecting it. The performance of the ensemble is compared to the individual classifiers used in the ensemble and also the other simple voting ensemble methods to verify whether the performance is better compared to them. The designed ensemble is currently implemented for sentiment analytics. This can also be used for other classification problems, where generalization is required for better results.

Keywords

Sentiment Analytics, Ensemble Method, Sensitivity, Specificity.

Full Text

     

T. Subbulakshmi ¹, S. Mercy Shalinie ¹, A. Ramamoorthi ¹

Affiliations
1 Department of Computer Science and Engineering, Thiagarajar College of Engineering, Tamil Nadu, IN

Abstract

Masqueraders are a category of intruders who impersonate other people on a computer system and use this entry point to use the information stored in the systems or throw other attacks into the network. This paper focuses on Ensemble Design of a Masquerader Detection System using Decision trees and Support Vector Machines for classification with two kernel functions linear and linear BSpline. The key idea is to find out specific patterns of command sequence that tells about user behaviour on a system, and use them to build classifiers that can perfectly recognize anomalous and normal behaviour. Real time truncated command line data set collected from a debian Linux server is used for performance comparison of the developed classifiers with the standard truncated command line data set of Schonlau[4]. The results show that Ensemble Design of Masquerader Detection Systems is much faster than individual Decision trees or Support Vector Machines.

Keywords

Masquerader Detection, Support Vector Machines, Decision Trees, Truncated Command Sequences.

Full Text