JOURNAL OF COMPUTERS, VOL., NO., NOVEMBER 009 Evaluatg the Network ad formato System Securty Based o SVM Model Shaome Yag Ecoomcs ad Maagemet Departmet, North Cha Electrc Power Uversty, Baodg Cty, Cha Emal: yagshaome@.com Dogle Lu College of Busess, Agrcultural Uversty of Hebe, Baodg Cty, Cha E-mal: doglelu@.com Zhb Lu Ecoomcs ad Maagemet Departmet, North Cha Electrc Power Uversty, Baodg Cty, Cha Emal: luzhb@.com Abstract Wth more ad more e-busess operatg o le, the etwork ad formato securty problem s ot oly cofed to the cocept of etwork techcal systems, but at the same tme has specal sgfcace beyod etwork tself, whch s ecoomc tes, cultural trasmsso meda, the socal operato ad maagemet way ad so o. Based o the aalyss of the securty mportace ad securty evaluato preset, ths paper establshes a securty evaluato system ad descrbes the evaluato mechasm based o SVM algorthm ad model. The securty evaluato of etwork eterprses Bejg shows that the results gve by ths model are relable, ad ths method to evaluate the etwork ad formato system securty s feasble. dex Terms etwork ad formato system, securty, securty evaluato, SVM. NTHRODUCTON Wth the through computer applcatos, especally the creasg popularty of home computers, o the oe had, t hopes that may users ca share formato resources, o the other had, t also hopes that ca trasmt formato ad commucate wth each other amog the computers. The hardware ad software cofgurato of the persoal computers s relatvely low geeral, ad the fucto s lmted, therefore, t requres that the hardware ad software resources of the large ad super computer ad ther maagemet formato resources should be shared for the large umber of mcro-computers, order to take full advatage of these resources. For these reasos, promote the computer s etwork developmet; coect the dspersed computer to the etwork; ad make up the computer etwork. Network s the combato product of the moder commucatos techology ad computer techology. The so-called computer etwork, s a etwork system of the large scale ad strog fuctos, whch les wth the computer dstrbuted dfferet geographcal regos ad the specalzed exteral equpmet commucato crcut, so that may computers ca trasmt formato to each other expedetly, ad share the resources of the hardware, software, Data formato, ad so o. Popularly, the etwork s the computer collecto whch les wth the cable, telephoe les, or wreless commucatos, ad other. formato system s defed techcally as a terrelated compoets collecto through the formato gatherg, processg, storage ad dstrbuto supported the decso-makg ad cotrol orgazato. Securty evaluato s also sad rsk evaluato. The defto of securty evaluato s: measure ad predct the system securty through usg sythetcally the securty systems egeerg approach, cofrm the possblty ad order of severty of the system dager through the qualtatve ad quattatve aalyss, ad put forward the ecessary measures order to fd the lowest accdet rate, the smallest cdet loss ad the best securty vestmet returs. Wth the emergece of all kds of e-mal spread vruses, people attach mportace to the etwork securty. More ad more customers wat to have a clear uderstadg about the etwork ad formato systems securty whch beg used or wll be used of ther ow, but most of them lack the relevat kowledge, expertse ad resources, who ca t decde whether the cofdece level of the etworks ad formato systems securty s approprate, ad do t hope to rely etrely o the system developer ths regard, so they wat a thrd party to help them aalyze the etwork system securty, that s, coduct securty evaluato. Through evaluato, the users ca judge whether the etworks ad formato systems s adequate securty for ther ow applcatos, ad whether the cocealed securty rsks are acceptable. addto, order to esure atoal formato frastructure securty, the maagemet departmet also 009 ACADEMY PUBLSHER
JOURNAL OF COMPUTERS, VOL., NO., NOVEMBER 009 call the shots that securty evaluato s very ecessary for a varety of etwork ad formato systems, ad to determe the level of safety ad relablty. Clearly, securty evaluato has the mportat sgfcace for the etwork developmet. The etwork ad formato systems securty evaluato s that detfes, aalyzes ad evaluates the securty, detfes all of the securty factors the system ad the lkages betwee them. A comprehesve securty aalyss cludg: descrbe varous levels of the securty probablty ad fluece, gve uform measuremet crtero to evaluate the securty dcators of the factors, ad the gve the securty level throughout the system based o some evaluato methods. The calculato methods of the securty evaluato have qualtatve method ad quattatve method. The ma evaluato methods cludg: AHP, decso tree aalyss, Markov modelg ad so o, however, these methods are subject to stochastc factors the evaluato, ad the evaluato results are flueced by subjectve experece ad kowledge lmtatos easly, whch affect the accuracy ad objectvty of the evaluato results. ths paper, we evaluate securty level usg the method of combg the qualtatve wth the quattatve, ad overcome the subjectvty the evaluato through the applcato of self-adaptg regresso Support Vector Mache (SVM).. THE SECURTY EVALUATON NDEX SYSTEM Accordg to the atoal basc requremets of the etwork ad formato securty system, wth the practcal experece egaged formato securty maagemet work may years, we beleve that should follow the followg prcples whe formulate the system securty evaluato dex system: ()tegrate the laws ad regulatos of the state formato ad formato system securty. ()Satsfy the securty requremets of the users ad applcato evromet to the formato systems. ()Good maeuverablty ad mplemetato expedetly. () Smple, practcal, feasble ad ecoomc. A Ettes ad evromet securty dex The ettes ad evromet securty dex cludes te aspects: Wthout dagerous buldg wth 00m aroud the motor room (U ), dagerous buldg s that the places of exstg flammable, explosve, toxc gases, etc., such as the such as gas statos, gas ppeles ad so o; Motorg system (U ), that s, the motorg mplemetato facltes of the exteral evromet ad operatg evromet durg the system rug; Freproof ad waterproof measures (U ), freproof s that equpped wth automatc fre alarm system the computer room, or the fre-fghtg equpmet, emergecy plas ad related systems appled to the computer room; waterproof s that o water seepage ad leakage the computer room, for example, t eeds waterproof layer f havg water facltes o the upper room; Evromet motorg ad cotrol facltes the motor room (temperature, humdty ad clealess)(u ), temperature cotrol s that the computer room has ar-codtog, ad the temperature mataed at C to C; humdty cotrol s that the relatve humdty remaed at 0% ~ 0%, clealess cotrol s that the computer room ad equpmet should be kept clea ad health, take off shoes to the computer room, the doors ad wdows wth closed performace; Thuder proof measures (havg thuder proof devces, the good groudg)(u ); Stadby power supply ad owed geerators (U ); Usg UPS (U ); At-statc measures (adopt at-statc floorg, equpmet groudg s good)(u ); Specal crcut supply electrcty (separate t from ar-codtog ad lghtg electrcty)(u 9 ); Guard agast theft measures(u 0 ), someoe s o duty, stall at-theft securty doors the mport ad export, stall metal defeded equpmet to the wdow, equp wth the rado-cotrolled at-theft etworkg facltes the computer room. B Orgazato maagemet ad securty system dex The orgazato maagemet ad securty system dex cludes eght aspects: specalzed formato securty orgazatos ad full-tme formato securty persoel (U ), the formato securty orgazatos establshmet ad the formato securty persoel appotmet, whch must have a offcal documet of the relevat uts; Perfect formato securty maagemet rules ad regulatos (U ); Strct maagemet system about formato securty persoel provdg or trasferrg (U ); The maagemet system of equpmet ad data s comprehesve, ad whch s o the wall (U ); The detaled the maual ad the tegrty work records(u ); The emergecy treatmet pla(u ); The tegrty plas ad system of the formato securty trag (U ); The securty resposbltes s clear for all types of employee ad maagers, ad securty maagemet system s strct(u ). C Securty techology dex The securty techology dex cludes seve aspects: dsaster recovery techology coutermeasures (U 9 ); Separato measures of the developmet work ad operatoal work (U 0 ); Havg applcato busess ad system securty audt fucto (U ); Havg the system operato log (U ), that s the wrtte records of opeg ad shuttg dow the computer, ad the equpmet operato codto, etc.; Server backup measures (U ); At-hackg facltes (U ), cludg set up a frewall, have truso detecto ad other facltes; Computer vrus preveto measures (U ), cludg the software ad hardware products of prevetg ad elmatg the vrus, ad the regular upgrades. D Network ad commucato securty dex The etwork ad commucato securty dex cludes fve aspects: havg eye-catchg sgs where placed commucatos facltes (U ); Backup of mportat commucato les ad cotrol devces (U ); Ecrypto measures (U ); Securty audt trackg measures of systems rug (U 9 ); Access cotrol 009 ACADEMY PUBLSHER
JOURNAL OF COMPUTERS, VOL., NO., NOVEMBER 009 measures of the etworks ad formato systems (U 0 ), s that, dvde the system user's access based o the work ature ad the rak. E Software ad formato securty dex The software ad formato securty dex cludes fve aspects: access cotrol measures of operatg system ad database (U ); Damage preveto measures of applcato software ad formato system (U ); Motorg faclty of the database ad system state (U ); User detfcato measures (U ); Remote backup of system user formato (U ).. THE PRNCPLE AND MODEL CONSTRUCTON OF SELF- ADAPTNG REGRESSON SVM SVM s a ew mache learg method proposed by Vapk based o the statstcal theory learg, ad s also a regresso method wth the good geeralzato ablty, whch lkes the eural etwork, has the capacty of approxmatg ay cotuous lmted olear fucto, ad SVM method has may advatages that the eural etwork has ot. SVM s from the Optmal Hyperplae uder the crcumstaces of the lear separable. The so-called Optmal Hyper-plae, s such a separatg hyper-plae, whch wll ot oly be able to classfy correctly for all trag sample, but make the dstace (defed as the terval)largest where from the proxmate pot to the Hyper-plae the trag samples. The vector of the dstace Optmal Separatg Hyper-plae s called support vector. A The basc prcple ad learg process of selfadaptg regresso SVM Self-adaptg support vector regresso algorthm based o RBF kerel fucto, wth regard to o sestvty coeffcet ε, value ε = 0.D () Amog them, D s varace of the ose dstrbuto fucto (radom process), durg the terato, the o sestve coeffcet ε remas uchaged. the k-th terato, the wdth coeffcets of pushmet factor ad kerel fucto are recorded as C (k) ad σ (k), usg these parameters, callg SVM learg algorthm, we ca carry o regresso estmato for the trag sample, ad the calculate the relatve fttg error of the trag samples: ( k ) y f ( x ) E = () y The average relatve fttg error s: ME = ( k ) E ( k ) = The regresso relatve accuracy s regard as δ, ad the adjustmet step of the pushmet factor ad wdth coeffcet are regard as C ad σ. Before the trag, frst of all, we should arrage all samples a orderly maer, that s regardg a sample optoally as m dmeso sample x (x, x,, x m ), choose the sample whch has the smallest dstace to x as x from the remag samples except for x, ad the () choose the sample whch has the smallest dstace to x as x from the remag samples except for x ad x, the same toke, wth regard to x, choose the sample whch has the smallest dstace to x as x + from the remag samples except for x,x,,x, at last, we ca arrage all samples a orderly maer as x, x,,x, the sample x + s the adjacet pot of the sample x, ad the meda of the adjacet pots: md x = ( x + 0.( x+ x )) () The dstace betwee the sample adjacet pots: d = x x () + The dstace mea s: d = d () = For orderly sample set {x, y } (=,,, ; x R m ; y R), the regresso fucto s f(x), suppose regresso accuracy s δ>0. As to the sample pot x (=,,,-), f the dstace betwee the adjacet pots s ot too bg, that s d <d, ad the depedet varable betwee the two adjacet pots s ot too close, that s y + - y >δ, we check the adjacet meda value f(x md ) of the regresso fucto, f t s t betwee y ad y +, the the regresso fucto s too complcated, whch ca lead to excessvely fttg. B The partcular steps of self-adaptg regresso SVM algorthm The partcular steps of support vector regresso method based o parameter self-adaptg adjustmet are as follows: () Select the parameters tal value, the terato tal value k=0, the wdth coeffcet tal value s: (0) T σ = ( x x) ( x x) () = the formula, s the umber of samples; x s the mea of put vector, ad the x = x () = The pushmet factor tal value s: (0) C = (max( y ) m( y )) (9) At the same tme value C>0 ad σ <0 by experece, proporto factor β = (0, ). We ca carry o support vector regresso evaluato usg the parameters tal value, ad the obta the average fttg relatve error ME (0). () k=k+, adjust parameters C (k) ad σ (k), carry o support vector regresso evaluato based o the adjusted parameters, ad the obta he average fttg relatve error ME ( k ). Parameters adjustmet formulas are: ( k + ) ( k ) C = C + C ( ) ( ) σ k + = σ k + σ (0) () Check f the average fttg relatve error s to be reduced, that s judgg whether ME (k+ ) <ME (k) s 009 ACADEMY PUBLSHER
JOURNAL OF COMPUTERS, VOL., NO., NOVEMBER 009 teable, f teable, the go to step (); f ot teable, order C (k+) = C (k), σ (k+) = σ (k), ad the go to step (). () Adjust parameter steps C ad σ, the parameter step adjustmet formula are: C = β C σ = β σ () Determe whether the step s less tha the mmum level, f ot, go to step (); f so, the go to steps (). () Judge whether ME (k+) <δ s teable, f ot, go to step (); f so, the go to step (). ()Determe whether the regresso fucto s too complcated. Check the adjacet meda value f(x md ) of the regresso fucto, f the value s betwee y ad y +, f ot, the the regresso fucto s too complcated, whle the umber of teratos s t the maxmum, order C (k+) = C (k), σ (k+) = σ (k), ad the go to step (); f the regresso fucto s t complcated or the umber of teratos s the maxmum, the go to step (). () The terato termato, at ths tme the regresso estmato fucto s the fal result. V. APPLCATON EXAMPLES We take the securty evaluato based o etwork eterprses survey data Bejg as a example, ad select eterprses data as the sample, cludg the excellet level, the good level Ⅱ, the geeral level Ⅲ, the poor level Ⅳ, take - samples as the trag set, the four resdual samples as testg samples. order to ehace the speed ad accuracy, we treat all samples data dfferetally. Costruct a four levels SVM classfcato; use the RBF kerel fucto, the fucto wdth σ=0., value C=000, use Matlab. programmg, treat the four levels SVM classfcato wth the trag samples show table, the trag tme s less tha 0.09S. We test the traed SVM classfer through usg test samples, the test results are table. order to check the method fucto, frstly, we evaluate the test sample based o the fuzzy eural etwork (FNN) method, ad the result s the exact same as the self-adaptg regresso SVM; the desg a BP artfcal eural etwork (ANN) category, the put layer euros s, whch correspodg separately securty evaluato parameters, the output layer euros s, whch correspodg separately four securty evaluato levels, the hdde layer ode umber s 0. We tra ad test base o the same samples ad the same cofgurato computer, the results show that ANN classfer has a msjudgmet whch takes a Ⅲ level securty sample as Ⅱ, ad other classfcatos are correct, the ANN classfcato ecessary trag tme s.s, whch s much hgher tha the SVM classfer trag tme. TABLE. NETWORK AND NFORMATON SYSTEM SECURTY EVALUATON DATA V 0.9 0. 0. 0.9 0.9 0.9 0.0 0.9 0.9 0. 0. 0. 0. 0.9 0. 0. 0. 0.9 0. 0. 0. 0. 0. CONTNUED TABLE No Level U U U U 9 U 0 V 0.0 0.9 0. 0. 0. 0.9 0. 0.9 0. 0.9 0.9 0.99 0.9 0. 0. 0.9 0.9 0.9 0. 0. 0. 0.0 0.9 CONTNUED TABLE No Level U U U U U V 0. 0. 0. 0.9 0.9 0.9 0.9 0. 0. 0.9 0.9 0.9 0. 0. 0.9 0. 0. 0.9 0. 0. CONTNUED TABLE No Level U U U U 9 U 0 0. 0. 0. 0.9 0.9 0.9 0.9 0. 0.9 No Level U U U U U 0. 0. 0. 009 ACADEMY PUBLSHER
JOURNAL OF COMPUTERS, VOL., NO., NOVEMBER 009 9 0. 0. V 0.9 0. 0. 0. CONTNUED TABLE No Level U U U U U 0.9 0.0 0.0 0.9 0.9 0.9 0. 0.9 0.0 0. 0.9 0. V 0. 0. 0. -, -, -, -,-, -,-, -,-,-, TABLE. THE ACTUAL EVALUATON RESULTS COMPARED WTH THE NETWORK TRANNG RESULTS AND CLASSFCATON No SVM SVM SVM SVM V 0. 0.9 0. CONTNUED TABLE - - - No Level U U U U 9 U 0 0.9 0.9 0.9 0.9 0.9 0.9 0.9 0.9 - - - CONTNUED TABLE No Evaluato results FNN ANN 0.0 0. 0. 0.0 0. 0.0 V V V V 0. 0. 0.0 CONTNUED TABLE No Level U U U U 0.9 0.9 0.9 0.9 0.9 0.99 0.9 0.9 0.9 0. 0.9 0. 0. 0. 0.0 0. 0.9 0. 0. V 0. 0. 0. CONTNUED TABLE No Level U Decso V. CONCLUSONS SVM s a commo learg method based o VC dmeso theory of the statstcal learg theory ad the structure rsk mmze (SRM) prcple, through the lmted sample formato, whch ca explore the best compromse betwee the complexty ad learg ablty of the model, ad the, order to receve the best outreach capacty. Accordg to the aalyss, qure to the mpact whch dfferet parameters to SVM regresso fucto, ths paper put forward regresso SVM method based o parameter self-adaptg adjustmet, order to avod select the best parameters through the complcated cross-certfcato steps. By the applcato of self-adaptg regresso SVM the etwork ad formato system securty evaluato, make up for the shortcomgs whch the sample data s few, ad the overcome the defects whch the tradtoal eural etworks may coverge to the local mmum pots ad the etwork structure determato ca oly deped o experece, ehace the geeralzato ablty, so mprove the system covergece speed ad evaluato accuracy. Examples show that self-adaptg SVM s a effectve method to securty evaluato, ad the t wll be good developmet prospects other areas. 009 ACADEMY PUBLSHER
0 JOURNAL OF COMPUTERS, VOL., NO., NOVEMBER 009 ACKNOWLEDGMENT Ths research was supported by the Scetfc Research Foudato for Youg Teachers of North Cha Electrc Power Uversty. The tem No. s 000. Ths research was supported by the Phlosophy ad Socal Scece Research Topcs of Baodg Cty. The tem s Research o Curret Stuato ad Coutermeasures of Agrcultural formato Degree Based o New Rural Costructo Baodg Cty, ad the tem No. s 000. REFERENCES [] Jgwe Ta, Mejua Gao, The research ad applcato o artfcal eural etwork algorthm[m], Bejg sttute of techology press, Beg, 00,, pp.-9. [] Weqg Cheg, Ja Gog, Network Securty Evaluato [J], Computer Egeerg, February 00, pp: -. [] Zhem GUO, Xuelog HU, Hulag JANG, Securty Evaluato of Network ad formato System ad ts dex System S Research [J], moder electro techology Sep 00, pp:9-. [] Dogme ZHAO, Hafeg LU, Cheguag LU, Rsk assessmet of formato securty based o BP eural etwork[j], Computer Egeerg ad Applcatos, 00, (), pp: 9-. [] Yaopg Jag, Research o Space Network Securty Evaluato dex System Cha [J], maagemet world, 00,, pp: -. [] Zhhu Wag, Fuhua Shu, The mproved SVM model ad ts applcato o coal me securty evaluato system [J], Mg dustry securty ad evrometal protecto, 00,,pp:-,. [] Graham Fracs, Matthew Hto, Jacky Hollo way, et a. Best practce bechmarkg: a route to compettveess [J], Joural of Ar Trasport Maagemet, 999 (), pp.0-l. [] Jl Xu, Applcato of Support Vector Mache Water qualty Evaluato [J], Cha coutrysde water coservacy ad hydraulc power geerato, 00,,pp:- 9. [9] KEERTH S, CHH J, Asymptotc behavor of support vector maches wth gaussa kerel[j], Neural Computato, 00,,pp: -9. [0] CHAPELE O, VAPNK V N, Choosg multple parameters for support vector maches [J], Mache Learg, 00,, pp: -9. [] Brow, M, H. G. Lews, S. R. Gu, Lear spectral mxture models ad support vector maches for remote sesg, (submtted to) EEE Tras, Geoscece ad Remote Sesg. 000. [] Dog X ad Zhaohu W, Speaker recogto usg cotuous desty support vector maches, Electrocs letters, 00,(-), pp: 099-0. [] Bahlma C, B. Haasdok ad H. Burkhardt, O-le hadwrtg recogto wth support vector machesa kerel approach, Proceedgs of Eghth teratoal Workshop o Froters Hadwrtg Recogto, 00, pp: 9-. [] Km K., Jm ad K. Jug, Recogto of facal mages usg support vector mache, Proceedgs of the th EEE Sgal Processg Workshop o Statstcal Sgal Processg, Sgapore, 00, pp: -. [] Jufeg Gao, Wegag Sh, Jaxu Ta et al, Support vector maches based approach for fault dagoss of valves recprocatg pumps, Proceedgs of EEE Coferece o Electrcal ad Computer Egeerg, Caada, 00,, pp: -. Shaome Yag, was bor Hada Cty, Cha, ad graduated from the agrcultural uversty of Hebe 00, gaed the master's degree of maagemet. The author s major feld of study s the busess maagemet. Sce 00, she s always workg at the North Cha Electrc Power Uversty, Baodg Cty, Cha. Ad she has publshed more tha papers ad book. Such as the Electrc Power Eterprse Maagemet (Bejg: Chese Electrc Power Publshg Compay). Dogle Lu, was bor Baodg Cty, Cha, ad graduated from the agrcultural uversty of Hebe 00, gaed the master's degree of maagemet. The author s major feld of study s the maagemet. Sce 00, she s always workg at the agrcultural uversty of Hebe, Baodg Cty, Cha. Ad she has publshed more tha 0 papers ad books. Zhb Lu, was bor Luaa Couty, Cha, ad graduated from the agrcultural uversty of Hebe 00, gaed the master's degree of maagemet. The author s major feld of study s the formato maagemet. Sce 00, he s always workg at the North Cha Electrc Power Uversty, Baodg Cty, Cha. Ad he has publshed more tha 0 papers ad books. Such as the Research o the Advacg Frot Topc of Asset Valuato (Bejg: Chese Face ad Ecoomcal Publshg Compay). 009 ACADEMY PUBLSHER