You are on page 1of 6

An Intelligent Fuzzy Rule based Feature

Selection for Effective Intrusion Detection


Riyaz. B Ganapathy. S
School of Computing Science and Engineering, School of Computing Science and Engineering,
VIT University- Chennai Campus, VIT University- Chennai Campus,
Chennai, India Chennai, India
riyaz.b2016@vitstudent.ac.in sganapathy@vit.ac.in

Abstract- A challenging issue today in the field of data mining techniques such as feature selection and classification
confidentiality is attack detection. This is because of ever algorithms. At the end of the analysis, the intruders were
increasing number of devices which are associated with the identified and detected. Feature selection [1] is a method
system. From the machine learning field, many unsupervised which selects a subset from the given input dataset based on
learning and supervised learning algorithms have been utilized the given rules. Feature selection has three different methods,
to build the benefit of Intrusion Detection Systems (IDSs). To namely filter method, wrapper method and hybrid method.
train a classifier, the labelled examples were utilized from the Filter method select the variables by random in the model.
supervised learning model, but getting adequate labeled Wrapper method is a learning algorithm which is used to train
examples goes complicated and it requires the domain experts itself iteratively for selecting the subsets. Moreover, it also
knowledge for training the model. From the real world problems used to simplify the models and acquires less training time and
some unlabeled examples can be obtained easily . Contrasted reduces the over-fitting value. Feature ranking and selection
with supervised learning and unsupervised learning models, this [3] is a vital point of view in intrusion identification
paper proposes a new Fuzzy Rule and Information Gain Ratio frameworks for improving the performance of classifier and
based feature selection algorithm (FRFSA) for preferring the
also reducing the execution time. Moreover, the features
greatest number of features which are used to classify the records
which are not contribute anything for making a decision
process in the given classifier that to be removed for
as normal and attacks. Moreover, the existing classifiers called
upgrading the detection accuracy and to improve the speed of
SVM and LSSVM are used for effective classification. The NSL-
intrusion detection systems.
KDD dataset is used for evaluating the proposed approach by
conducting some experiments. The obtained outcomes from this Classification is an essential part in data mining
work, proved that the proposed work exceeds the performance techniques which is able to take a decision over the dataset.
measure when compared to the existing algorithms on Generally, classification algorithms can be classified into three
classification for feature selection. such as one class, two class and multiclass classification.
Moreover, these classifiers also can be categorized in two such
Keyword- Intrusion detection system; Feature selection; Data as supervised and unsupervised based on their working nature.
Preprocessing; Support vector machine Among them, the supervised classification algorithm classified
as namely k - nearest neighbor (KNN) [6], Support vector
I. INTRODUCTION
machine (SVM) [7] and Neural networks (NN) [8]. These all
Computer network security incorporates the set of rules kinds of classification algorithms manufacture the prototype
which are used to monitor and detect the unauthorized access. which confines another concealed case or occurrence with the
It also prevents the malicious activities such as modify and right label. Here, some precedence and the complications were
misuse the data, denial of network services and accessible identified with the supervised learning method with IDS
resources. This network security includes access control which has been accounted by numerous researchers. Some
mechanism, Intrusion Detection Systems (IDSs), specific supervised learning method was inadequacies the
authentication policies and firewalls and. Among them, the requirement for the labeled instances.
IDSs are most widely used in the world for identifying and
detecting the intruders in Computer Networks, Internet and Fuzziness indicates that the inaccuracy score between any
Cloud Networks. The IDS is a system which monitors and two linguistic terms that depends on the fuzzy membership
detects the network intruders for providing the better services function. Many researchers concluded that the fuzziness as a
to society. sort of uncertainty and inadequate data in their analysis.
Membership function has the real time interval between 0 and
Data mining is a technique for analysing the existing data 1. This real time value precisely described that whether the
from the different perspectives to generate new useful object is possibly belongs to a set or not. Here, a fuzzy
information. Data mining approaches are used to classify the inference framework is used to link the input and output
network dataset and the trace dataset (i.e. real-time dataset) for values by applying fuzzy calculations. The initial process is to
providing network security by identifying and detecting the fuzzify the contribution through some membership function,
intruders. The network data set can be analyzed by using data and bringing out a fuzzy set. Evaluating the obtained fuzzy set

978-1-5386-5853-6/18/$31.00 2018
c IEEE 206

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on August 30,2020 at 07:30:38 UTC from IEEE Xplore. Restrictions apply.
by a set of guidelines which setting it up for the next process with a enormous number of characteristics. The objective of
called defuzzification technique. At the end, a crisp dataset is their model is to perform well in execution process and to
obtained from the defuzzification process as a output, that is decrease the number of wrapper assessments. Their algorithm
for the given classification issue the last outcome is the label switched over from filter method to wrapper method for subset
of the given input contribution which being assessed [23]. selection for accomplishing the performance improvement.
Likewise, the strategy was tested on eleven high dimensional
The rest of the paper is arranged as follows: Section II
informational sets utilizing diverse classifiers.
provides a brief outline about the related works in the
direction of intrusion detection system. Section III explains the Ganapathy et al [13] presented an unique weighted fuzzy
overall system proposed architecture. Section IV described in C- Means clustering algorithm based on an immune genetic
detail about the proposed model. Section V shows the algorithm for effective intrusion detection. Their algorithm
proposed experimental results. Section VI gives the performs the data clustering process which is useful for
conclusions and future directions. identifying the anomalies in the network and also classified
the data as misuse and anomaly. Their algorithm produces
II. RELATED WORK
better classification accuracy when compared to the other
In the past, several researchers have been worked in the existing models. Ganapathy et al [15] proposed a novel
direction of this proposed work[4,5,9,14,16]. From the intelligent agent-based IDS that uses the Fuzzy Rough set
existing works, Aamir et al [2] proposed an unique fuzziness based SVM and the Fuzzy Rough Set based outlier detection.
based learning approach which is semi-supervised model by In their work, the authors have produced two diverse of
using the unlabeled examples that assisted with the supervised intelligent specialists that were to be specific feature selection
learning method for improving the IDSs classifiers operator for choosing the appropriate feature dataset by using
performance. In their approach, a single hidden layer feed- fuzzy rough sets method and the decision making
forward neural network was trained to produce output as a administrator for generating final decision. In addition, they
fuzzy membership function and the level classification such as also demonstrated that fuzzy rough set based SVM for
low fuzziness, mid fuzziness and high fuzziness over the identifying the anomalies. Finally, they evaluated their model
unlabeled records using fuzzification process. Their approach by conducting various experiments by using KDD Cup 99
was retrained after incorporating the level categorization dataset and proved that their model improved the overall
separately into the original training dataset. They have used detection accuracy and reduction in false alarm rate.
the NSL-KDD dataset for evaluating their model and proved Mohammed et al [19] introduced a novel IDSs called
that the unlabeled records belongs to the small fuzziness and Least Square Support Vector Machine based IDS (LSSVM-
the huge fuzziness groups that are contributed more for IDS) which uses the newly proposed feature selection
improving the performance of classifier when compared to the algorithm called MMIFS that handles the linearly dependent
existing classifiers like Naive Bayes classification, Decision and nonlinearly dependent data features. The proposed feature
Tree, SVM and random forests method. selection algorithm selects the best count of features that are
Amiri et al [10] introduced a distinct feature selection used to enhance the classification accuracy performance in the
algorithm called modified mutual information feature proposed IDS. Vitali et al [20] introduced a new feature
selection (MMIFS) algorithm for effective feature selection. In selection method which truncate the quantity of features and
addition, they conveyed in their work about the feature instances in the trained model collection by without largely
examination process by using two feature selection strategies affecting the accomplished accuracy of IDSs. Lastly, they
such as linear correlation coefficient method and forward calculated their method using four distinctive rule-based
feature selection algorithm. They also investigated by using classifiers that were tested over the real scan and the
both non-linear and linear measures for the feature selection backscatter data that were gathered by a network telescope.
process. They used KDD’99 cup dataset. Analyses were Akashdeep et al [21] proposed an intelligent IDSs which
performed this new method on the windows platform and the function with the selected number of features. Moreover, the
result demonstrated that MMIFS is major viable in identifying proposed framework selects the features by utilizing the
the probe attacks and R2L attacks while FFSA execution was information gain ratio concept and the correlation values. In
great in detecting the typical profile attack, DoS attacks and addition, features were ranked by utilizing the information
U2R attacks. gain and correlation values and also combined, after that by
Sung and Mukammala [11] introduced a new feature utilizing a properly designed method. Moreover, their
ranking technique which is used to enhance classification technique also uses the pre-processing activity for eliminating
accuracy of their classification model. Their classification the redundant data and the irrelevant data from the standard
model has the SVM and the neural network system for benchmark dataset for improving the resource utilization and
performing the classification of attacks. They have gained also reduces the time complexity. Sannasi et al [22] introduced
better detection accuracy in entire types of attack detection. a new IDS for detecting the various types of attacks by
Barmejo et al [12] proposed a new feature selection technique utilizing the newly proposed feature selection algorithm called
which manages the subset determination in the given datasets intelligent Conditional Random Field (CRF) based feature
selection algorithm for selecting the optimal number of

International Conference on Recent Trends in Advanced Computing (ICRTAC-CPS 2018) 207

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on August 30,2020 at 07:30:38 UTC from IEEE Xplore. Restrictions apply.
features and the newly proposed Intelligent Layered Approach Fig. 1. Proposed System Architecture
for identifying the different types of attacks effectively.
Moreover, the proposed algorithm is used to enhance the
proposed classifier performance. In addition, their classifier
uses an intelligent agent for making effective decision over the The proposed model consists of five major modules such as
intrusion dataset. NSL-KDD dataset, Data Collection Module, Data
preprocessing, Fuzzy system, Classification module and
Enamul et al [24] presented the LS-SVM based intrusion Result dataset. The data collection module collects the
detection for classifying the large datasets. For huge number necessary features from the standard NSL-KDD dataset and
of datasets, it is important to reducing the dataset dimensions forwards them to the preprocessing module for performing
and classify the given dataset for identifying intrusion by pre-processing activities. The preprocessing module is
using some classifiers. For that, the required size should be responsible for cleaning the dataset using the various
determined by describing the huge dataset characteristics and subcomponents such as data cleaning, data integration, data
then divide the complete dataset into smaller subgroups and transformation, data reduction and data discretization.
then a sample was selected from the subgroups using optimum
allocation method. The selected sample was given as input to Finally, it selects the most contributed features from the
the LS-SVM to identify different attacks in the intrusion datasets by using the proposed feature selection algorithm.
detection system. OA-LS-SVM is the proposed method which The existing classifiers such as SVM and LSSVM are used in
identifies the testing and training dataset size depends on the the classification module for classifying the records as
variability of data. Then select the sample from the subgroups ‘normal’ and ‘attack’.
of testing and training dataset that depends on the weakness by
providing a dependable means of data reduction. IV. PROPOSED WORK

III. SYSTEM ARCHITECTURE This paper proposes a novel feature selection algorithm
called Fuzzy Rule and Information Gain Ratio based Feature
Fig.1 presents the overall proposed architecture of the Selection Algorithm (FRFSA) for selecting most favourable
intrusion detection model. number of contributing features which are useful for effective
classification and also uses the two available classification
algorithms namely SVM and LSSVM for classifying the
intrusion dataset as normal and attack. Here, the standard
benchmark dataset called NSL-KDD dataset is used for
evaluating the proposed model.
NSL-KDD dataset is the refined variant of KDD’99 cup
dataset. From the 41 features the redundant records were
removed and only sufficient quantity of records are available
for training and testing the dataset. NSL-KDD is used for
evaluating the effectiveness of proposed algorithm for
detecting the intrusions. Each of the network traffic record in
the obtained dataset contains 41 features and 1 class label.
From the characteristics of features, the attacks in the dataset
classified into denial of service attacks, probe attacks, root to
local attacks and user to remote attacks.
A. Feature Selection

This subsection explains in detail about the proposed


FRFSA for choosing the optimal number of most useful
features. The following steps of the proposed algorithm are
explained as below:
Fuzzy Rule and IGR based feature selection algorithm

Input: NSL-KDD Dataset


Output: List of selected Features
Step 1: Read the NSL-KDD data set.
Step 2: Calculate Information Gain (IG) value by using:

208 International Conference on Recent Trends in Advanced Computing (ICRTAC-CPS 2018)

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on August 30,2020 at 07:30:38 UTC from IEEE Xplore. Restrictions apply.
ሺೌషŸሻ;
ି
Ƀ=ͳ െ ݁ మ഑మ (1)

where a is for membership calculation of the real traffic


Inf (D) = -
୫ and â is midpoint predicted traffic and σ is the standard
ˆ”‡“ሺŒǡ ሻ deviation parameter for the threshold value. The result dataset
෍ቈ ቉ Ž‘‰ʹሾˆ”‡“ሺŒǡ ሻȀȁȁሿ
ȁȁ features represents the network behaviour such as normal and
୨ୀଵ
abnormal attacks.
Inf (F) = σ୫
୧ୀଵሾȁ ‹ȁȀȁ ȁሿ ‫ˆ‹ כ‬ሺ ‹ሻ Using this Gaussian membership function we obtained
IG (Ai) = [Info (D) – Info (F) / Inf (D) + Inf (F)]*100 some membership values as output and summing up those
values to a single value. The defuzzification function checks
Step 3: Apply fuzzy rules for removing unwanted features the obtained sum value to the general threshold value which
Step 4: Calculates the mean value of IG for the selected was defined as Ѓ. The abnormality degree is defined if any
features (SF) anomaly behaviour in the network data. In this proposed work,
the threshold value classifies the normal and abnormal attacks
Step 5: If IG (SFi ) < Threshold then in the given dataset.
Remove SFi from the selected feature list (SF)
First Rule: If σହ௜ୀ଴ ߞ݅ < Ѓ , then Normal (2)
Else
Apply CRF for selecting the features Second Rule: If σହ௜ୀ଴ ߞ݅ ≥ Ѓ, then Abnormal (3)

Step 6: Select the feature set according to the conditional The anomaly detection has the degree of uncertainty and the
probability. threshold value represents the detection rate and effectiveness
Step 7: List the selected features of the classification.

where, C. Classification
D represents the dataset, Cj represents number of classes, This subsection describes in detail about the classification
Fi represents maximum number of non-zero values chosen by module of the proposed intrusion detection model. The
the agent. proposed classification module uses a hybrid classifier which
consists of two existing classification algorithms called SVM
The proposed feature selection algorithm calculates the IG [17] and Least Square Support Vector Machine (LSSVM)
value for all records of the input dataset. Then, remove the [18]. Here, the supervised algorithm strategy is utilized for
unwanted or less contributed features from the dataset by handling the classification and regression issues. Here, the
using the proposed fuzzy rules. The fuzzy rules can be framed SVM uses an expansive number of patterns and the LSSVM
in the form of IF…THEN with the consideration of the uses the revised version of the SVM patterns that are available
various information gain values of all the features. After already. In LSSVM, understands a linear equation which is
removing the unwanted or less contributed features from the from the optimization stage and thus the procedure is also
dataset, mean value of information gain is determined for all simplified in this module. Besides, the LSSVM was powerful
the useful features in the given dataset. Then checks whether since it avoids the neighborhood minima in the SVM issues
the information gain value of the selected feature is less than that are utilized by LSSVM classifier is utilized to recognize
the threshold or not. If the information gain value is less then, ordinary and attacks information.
removes the particular feature from the selected features list.
Otherwise, the most contributed features have been identified
V. RESULTS AND DISCUSSION
and selected by applying conditional random probability.
Finally, list the most useful and contributed features as result. The proposed feature selection algorithm has been carried
These selected featured record sets only will be used for out by using JAVA language and also uses the existing SVM
testing in the classification module of the proposed IDSs. and LSSVM classifiers codes which are developed in same
programming for evaluating the performance. The
B. Using Fuzzy Rule for IDS performance of the proposed FS algorithm with SVM has been
A membership degree of an element was obtained from the tested with WEKA tool. The performance of the classifiers
various parameters of the fuzzy membership function. along with the proposed FS algorithm is tested by using
Gaussian membership function is used in this proposed work. tenfold cross-validation. Table 1 shows the selected list of
features by the proposed FS algorithm. Here, the proposed
feature selection algorithm selects only 14 features from 41

International Conference on Recent Trends in Advanced Computing (ICRTAC-CPS 2018) 209

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on August 30,2020 at 07:30:38 UTC from IEEE Xplore. Restrictions apply.
features that are available in the benchmark dataset because of
the usage of proposed intelligent fuzzy rules and the
information gain ratio value. The selected features are showed
in the table as given below:

TABLE I. LIST OF SELECTED FEATURE

S.No. Name of the features


1 Duration
2 Protocol_ type
3 Service
Fig.3 Attack Detection Accuracy Analysis
4 Src_ bytes
From fig.2 and fig.3, the performance of the proposed FS
5 Num_ failed_logins
algorithm performs well was observed with the existing
6 Logged_in classifier LSSVM than SVM in DoS attack detection. The
7 Num_ compromised model was trained and tested by the NSL-KDD data for
8 Num_ file_creations identifying the DoS, R2L, U2R, Probe attacks was performed
and the accuracy was determined. The Dos attack detection
9 Num_ access_files
accuracy is good when it was contrasted to the detection
10 Is_ host_login accuracy of other attacks like R2L, Probe and U2R. This is
11 Is_ guest_login due to the fact that the use of intelligent agent and IGR on the
12 Dst_ host_srv_ serror_ rate feature selection algorithm. The accuracy was improved
13 Dst_ host_rerror_ rate
because the attributes which are given correctly for making the
good decision only are considered by the proposed model.
14 Dst_ host_srv_rerror_ rate
Fig. 4 shows the proposed feature selection algorithm
Fig.2 and fig.3 shows the attack detection accuracy performance analysis with various classification algorithms
analysis for the different types of attacks such as Probe, DoS, such as Naïve Bayes, Naïve Bayes with the proposed feature
U2R and R2L attacks by the proposed intrusion detection selection, SVM, SVM with the proposed feature selection,
system. Here, we have conducted five different experiments LSSVM and the LSSVM with the proposed feature selection.
such as E1, E2, E3, E4 and E5 with the distinct set of records
for demonstrating the performance of the proposed FS
algorithm on different classifiers such as SVM and LSSVM.

Fig. 4. Performance Analysis

Fig.2 Attack Detection Accuracy Analysis From fig.4, it shows that the performance of the proposed
feature selection algorithm performs well when it is contrasted
with the existing algorithms that are using full features. Even
though, it shows better performance on LS-SVM when
compared to other classifier methods. This is as a result of the
case that the use of intelligent agent and IGR on feature
selection process.

210 International Conference on Recent Trends in Advanced Computing (ICRTAC-CPS 2018)

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on August 30,2020 at 07:30:38 UTC from IEEE Xplore. Restrictions apply.
VI. CONCLUSION AND FUTURE WORK [17] C. Cortes, V. Vapnik, “Support vector networks”, Machine Learning.
20,pp. 1–25, 1995.
[18] J. Suykens, L Lukas, DP Van, MB De, J Vandewalle, “Least squares
In the proposed work, a novel fuzzy rule and IGR based support vector machine classifiers: a large scale algorithm”, in
feature selection algorithm has been implemented for selecting Proceedings of the European Conference on Circuit Theory and Design,
the most contributed features which are used to strengthen the pp. 839–842, 1999.
classifiers performance in terms of classification accuracy and [19] A. Mohammed , Xiangjian He, Priyadarsi Nanda, Zhiyuan Tan,
"Building an Intrusion Detection System Using a Filter-Based Feature
reduction in time. The experiments were handled by using the Selection Algorithm", IEEE Tran. on Computers, Vol.65, No.10, pp.
defined NSL-KDD dataset for calculating the performance of 2986- 2998, 2018.
the proposed work. Experimental results proved that the [20] Vitali H S, Osvaldo A Pérez-García, Raudel Hernández-León, Jan van
proposed method exceeds the decision making process than den Berg, Christian Doerr, "A data reduction strategy and its application
on scan and backscatter detection using rule-based classifiers", Expert
the existing methods. Future work in the proposed direction Systems with Applications, Vol. 95, pp. 272-279, 2018.
could be the introduction of new intelligent agent-based [21] Akashdeep, Ishfaq Manzoor, Neeraj Kumar, "A feature reduced intrusion
classifiers for effective decision making. detection system using ANN classifier", Expert Sys. with App., Vol.88,
pp. 249-257, 2017.
REFERENCES [22] Sannasi G, P. Vijayakumar, Palanichamy Yogesh, Arputharaj Kannan,
"An Intelligent CRF Based Feature Selection for Effective Intrusion
[1] S. Ganapathy, K. Kulothungan, S. Muthurajkumar, Vijayalakshmi, P. Detection", International Arab Jou. of Info. Technology, Vol.13, No.1,
Yogesh, and A. Kannan, “Intelligent Feature Selection and pp. 44-50, 2016.
Classification Techniques for Intrusion Detection in Networks: A [23] Anderson Hiroshi H, Luiz Fernando C, Lucas Dias Hiera Sampaio,
Survey,” Comm.. Network., Eurasip journals, Springer, Vol. 2013, No. Taufik Abrão, Mario Lemes Proença,“Network Anomaly Detection
1, pp. 271, 2013. System using Genetic Algorithm and Fuzzy Logic”, Expert Sys. with
[2] R. Aamir, R. Ashfaq, X. Wang, J. Zhexue, H. Abbas, and Y. He, Applications, Vol. 92, pp. 390-402, 2018.
“Fuzziness based semi-supervised learning approach for intrusion [24] K. Enamul, Jiankun Hu, Hua Wang, Guangping Zhuo, “A novel
detection system,” Info. Sci. (Ny), vol. 378, pp. 484–497, 2017. statistical technique for intrusion detection systems”, Future Generation
[3] M. Liu and D. Zhang, “Feature selection with effective distance", Neuro Comp. Systems, Vol. 79, pp. 303-318, 2018.
comp., Elsevier, http://dx.doi.org/10.1016/ j. neu.com.2015.07.155,
2016.
[4] I. Manzoor and N. Kumar, “A feature reduced intrusion detection system
using ANN classifier,” Expert Syst. Appl., vol. 88, pp. 249–257, 2017.
[5] KDDCup 1999 Data, Available: http://kdd.ics.uci.edu/ databases/
kddcup99/ kddcup99.html, 2015.
[6] Y. Liao , V. Vemuri , :Use of k-nearest neighbor classifier for intrusion
detection”, Comput. Secur. 21, vol. 5 , pp. 439–448,2002.
[7] S. Mukkamala , Sung , “Detecting denial of service attacks using
support vector machines”, Proceedings of the Twelfth IEEE
International Conference on Fuzzy Systems, 2003 .
[8] S. Mukkamala , G. Janoski , A. Sung , “Intrusion detection using neural
networks and support vector machines”, Proceedings of the 2002
International Joint Conference on Neural Networks. IJCNN02 (Cat.
No.02CH37290), vol. 2, pp. 1702-1707, 2002.
[9] F. Wang and J. Liang, “An efficient feature selection algorithm for
hybrid data,” Neuro computing, Elsevier, vol. 193, pp. 33–41, 2016.
[10] F. Amiri, , M R Yousefi. ,C Lucas , A Shakery , & N Yazdani, “Mutual
information based feature selection for intrusion detection”, Network
and Computer Application, 34 , 1184–1199,2011 .
[11] A H Sung, and S Mukammala, “Feature selection for intrusion detection
using neural network and support vector machine”, Transportation
Research Record: Jou. of the Transportation Research Board, 1822, pp.
1–11, 2003.
[12] P Barmejo ,L Ossa, J A Gamez and J M Puerta, “Fast wrapper feature
subset selection in high dimensional datasets by means of filter re
ranking”, Jou. of Knowledge Based Systems, vol. 25 ,pp. 35–44, 2012.
[13] S Ganapathy, K.Kulothungan, P Yogesh, A Kannan, “A Novel
Weighted Fuzzy C-Means Clustering based on Immune Genetic
Algorithm for Intrusion Detection”, Procedia. Engg, ScienceDirect, vol.
38, pp. 1750–1757, 2012.
[14] P B Thiago Vieira, F Danilo, Joao Paulo, P Edison, ”Model order
selection and eigen similarity based framework for detection and
identification of network attacks”, Jou. of Network and Computer Appl.,
Vol.90, pp.26-41, 2017.
[15] N. Jaisankar, S. Ganapathy, P. Yogesh, A. Kannan, and K. Anand, “An
Intelligent Agent Based Intrusion Detection System Using Fuzzy Rough
Set Based Outlier”, Soft Comp. Tech. in Vision Sci, vol.395, pp. 147–
148, 2012.
[16] F. Amato, N. Mazzocca, E. Vivenzio, D. Ingegneria, T. Informazione,
and F. Moscato, “Multilayer Perceptron : an Intelligent Model for
Classification and Intrusion Detection,” International Conference on
Advanced Info. Networking and App. Workshops, IEEE,pp.686-
691,2017.

International Conference on Recent Trends in Advanced Computing (ICRTAC-CPS 2018) 211

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on August 30,2020 at 07:30:38 UTC from IEEE Xplore. Restrictions apply.

You might also like