Record Details

DATAMINING FRAMEWORK FOR NETWORK INTRUSION DETECTION USING EFFICIENT TECHNIQUES

Shodhganga@INFLIBNET

View Archive Info
 
 
Field Value
 
Title DATAMINING FRAMEWORK FOR NETWORK INTRUSION DETECTION USING EFFICIENT TECHNIQUES
—
 
Contributor Dr.G.Jagajothi
 
Subject —
 
Description An Intrusion Detection System (IDS) checks the connection records, traffic control
newlinepackets for identifying the intrusion or attacks. The amount of records generated by a network
newlineis huge in quantity. Features are extracted from the records by the IDS and then classified to
newlineidentify the record/connection as attack or normal traffic. To facilitate the machine learning
newlinemethods used for classification, it is feasible to reduce the dimension of the feature. In this
newlineresearch it is proposed to develop general and systematic methods for classifying intrusion
newlinedetection. The key ideas are to use data mining techniques to discover consistent and useful
newlinepatterns of system features that describe network behaviour, and use the set of relevant system
newlinefeatures to recognize anomalies and known intrusions.
newlineThe problem with current IDS is that they are tuned specifically to detect known service
newlinelevel network attacks. Attempts to expand beyond this limited realm typically results in an
newlineunacceptable level of false positives. At the same time, enough data exists or could be collected to
newlineallow network administrators to detect these policy violations. Unfortunately, the data is very
newlinehigh, and the analysis process so time consuming, that the administrators don t have the resources
newlineto go through it all and find the relevant knowledge.
newlineIn this research, existing data mining algorithms are investigated for classifying normal and
newlineabnormal traffic and new algorithms are proposed. The UDP data streams from the KDD 99 data
newlineset extracted and create a multi class dataset specifically highlighting the different intrusion threats
newlinecommon to UDP data streams. The signatures extracted from the dataset were used to check the
newlineclassification accuracy of Naïve Bayes Algorithm, Random Tree and Neural networks. Principal
newlineComponent Analysis (PCA) and Fisher Score are evaluated for feature reduction in this work.
newlineInitial experiments were conducted without any dimension reduction of the feature set and second
newlineset of experiments were conducted on a reduced dataset by applying PCA. The PCA improves
newlineclassification accuracy by 1.66%.
newlineInspired by the human immune system, Artificial Immune Systems (AIS) are algorithms
newlineand mechanisms which are self-adaptive and self-learning classifiers capable of recognising and
newlinev
newlineclassifying by learning, long-term memory and association. Unlike other human system inspired
newlinetechniques like genetic algorithms and neural networks, AIS includes a range of algorithms. The
newlineAIS has different algorithms implementing different properties in different cells. Experimental
newlineresults show that AIS achieves higher classification accuracy of 99.35% when compared to
newlinegenetic algorithm which achieved an accuracy of 98.65%.
newlineIn the final stage, a framework for IDS with genetic search method for feature selection,
newlineand proposed AIS based on apoptosis is used for classification is proposed. The Artificial Immune
newlineSystem (AIS) protects a complex system against malicious defects, thereby achieving its survival
newlinepolicy by an extension of the concept of organization of multi-cellular organisms to the
newlineinformation systems. The main features of AIS are self-maintenance, distributed and adaptive
newlinenature of the computational systems. Although AIS is used for pure optimization, it can be made
newlineto work; this is probably the missing point. AIS are powerful when a population of solutions is
newlineessential either during the search, or as an outcome. The concept of matching is involved in
newlinefinding the solution. Since AIS are evolutionary algorithms, they are more suitable for problems
newlinethat change over time, and need to be solved again and again, rather than one-off optimizations.
newlineExperimental results demonstrate the efficiency of the proposed AIS based on apoptosis method.
newlineIt is observed that the proposed AIS based on apoptosis method without GA based feature
newlineselection classification accuracy 99.79%, and with GA based feature selection classification
newlineaccuracy 99.88% is achieved.
newline
—
 
Date 2017-07-27T06:34:33Z
2017-07-27T06:34:33Z
10-6-2009
—
23-12-2016
 
Type Ph.D.
 
Identifier http://hdl.handle.net/10603/163535
 
Language English
 
Relation —
 
Rights university
 
Format —
—
CD
 
Coverage —
 
Publisher Thanjavur
Periyar Maniammai University
Department of Computer Science and Engineering
 
Source University