Research Article Multilevel Association Rule Mining for Bridge …downloads.hindawi.com/journals/aaa/2014/278694.pdf · 2019-07-31 · Research Article Multilevel Association Rule

Research ArticleMultilevel Association Rule Mining for Bridge ResourceManagement Based on Immune Genetic Algorithm

Yang Ou1 Zheng Jiang Liu1 Hamid Reza Karimi2 and Ying Tian3

1 College of Navigation Dalian Maritime University Dalian 116026 China2Department of Engineering Faculty of Engineering and Science The University of Agder 4898 Grimstad Norway3 College of Engineering Bohai University Jinzhou 121000 China

Correspondence should be addressed to Ying Tian tianyingjzh163com

Received 31 December 2013 Revised 22 January 2014 Accepted 22 January 2014 Published 27 February 2014

Academic Editor Shen Yin

Copyright copy 2014 Yang Ou et alThis is an open access article distributed under the Creative Commons Attribution License whichpermits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

This paper is concerned with the problem of multilevel association rule mining for bridge resource management (BRM) which isannounced by IMO in 2010The goal of this paper is tomine the association rules among the items of BRM and the vessel accidentsHowever due to the indirect data that can be collected which seems useless for the analysis of the relationship between items of BIMand the accidents the cross level association rules need to be studiedwhich builds the relation between the indirect data and items ofBRM In this paper firstly a cross level coding scheme for mining the multilevel association rules is proposed Secondly we executethe immune genetic algorithm with the coding scheme for analyzing BRM Thirdly based on the basic maritime investigationreports some important association rules of the items of BRM are mined and studied Finally according to the results of theanalysis we provide the suggestions for the work of seafarer training assessment and management

1 Introduction

With the development of shipping science improvementof vessel technology and the rising trend of multinationalmanning of ships maritime safety and marine pollutionprevention have been the hot spots in study of the maritimefield by all the countries Against this background the focusesof maritime safety research are mainly on two aspects One isthe reasonable use of marine technology resources which istermed as the seamanship The other is the proper trainingof the communication and cooperation in the sailing teamwhich is named as the teamwork in the ship

In fact the good seamanship and the excellent teamworkin the deck crew management of the ship are both the mostimportant factors to avoid the vessel accidents The previousresearches about the ship accidents show that more than 80accidents are caused by human errors which are inseparablewith the level of seamanship and the teamwork Thereforein 2010 the IMO announced the sweeping amendments ofthe STCW convention named STCW convention Manila

amendments In the convention marine resource manage-ment ability which is the collectivity name of seamanship andthe teamwork becomes one of the mandatory requirementsand both bridge resource management (BRM) and engineresource management (ERM) are proposed formally How-ever STCW Manila amendments only explain the marineresource management qualitatively So far a few literaturesfocus on the data analysis between the vessel accidents anditems of BRM Paper [1] promoted the rating model forseafarer based on BRM but it gave the importance of itemsof BRM by the expert questionnaire without data analysisTherefore it is necessary to find some associations which canreflect the deep inducements of accidents from the collectedmaritime accident data to guide the seafarer training

Generally speaking a large amount of maritime acci-dent data can be collected from the actual vessel accidentassociation rule mining which is a kind of data miningis a feasible method to analyze this data During recentyears many researches about the association rule miningare proposed [2ndash8] Literature [2] is proposed to integrate

Hindawi Publishing CorporationAbstract and Applied AnalysisVolume 2014 Article ID 278694 8 pageshttpdxdoiorg1011552014278694

2 Abstract and Applied Analysis

classification and associationmining techniques according tothe target of discovery are not predetermined in associationrule mining The integration is done by focusing on mininga special subset of association rules called class associationrules (CARs) An efficient algorithm is also given for buildinga classifier based on the set of discovered CARs In [3]an efficient representation scheme called general rules andexceptions is proposed This paper focused on using therepresentation to express the knowledge embedded in adecision tree In this representation a unit of knowledgeconsists of a single general rule and a set of exceptions Thisscheme reduces the complexity of the discovered knowledgesubstantially It is also intuitive and easy to understandAuthors in literatures [4 5] deal with the algorithmic aspectsof association rulemining and they explain the fundamentalsof association rule mining and moreover derive approachesPaper [6] presents effective techniques for pruning the searchspace when computing optimized association rules for bothcategorical and numeric attributes However to mine theassociation rules effectively it needs to tackle the problemson the very large search space of candidate rules during therule discovery process Some intelligent searching methodsare used in the rule discovery process such as literatures[7 8] The paper [7] seeks to generate large item sets ina dynamic transaction database using immune clone algo-rithm where intratransactions intertransactions and dis-tributed transactions are considered for mining associationrules The paper [8] proposes an approach which is basedon artificial immune systems for mining association ruleseffectively for classification Instead of massively searchingfor all possible association rules this approach only findsa subset of association rules that are suitable for systemeffective in an evolutionary manner However producing toomany rules have been regarded as a major problem withmany data mining techniques The large number of rulesmakes it very hard for manual inspection of the rules togain insight of the domain Many data mining methodsmentioned above typically generate a list of rules with nofurther organization It is like writing a book or a paper byrandomly listing out all the facts and formulas The resultingbook will be difficult for anyone to read because one cannotsee any relationships among the facts and formulas In papers[9 10] they proposed a technique to intuitively organize andsummarize the discovered rules With this organization thediscovered rules can be presented to the user in the way asthe thinking and talking about knowledge in the daily livesA common strategy for organizing the large number of rulesis by hierarchy In such an organization the rules are summa-rized into topics and presented at different levels of detailsThe main advantage of the hierarchical organization is that itallows the user to view the problem from general to specificand to focus hisher attention on those interesting aspectsPapers [9 10] manage the complexity of the discovered rulesin a hierarchical fashion but these methods are too complexto deal with the maritime accident data In our paper weuse the hierarchical thinking organizing lots of items intomultiple levels by the cross level coding Additionally theimmune genetic algorithm is used in the rulemining processBy this way the major association rules which reflect deep

inducements of accidents will be mined effectively in anevolutionary manner

This paper mainly researches BRMThrough the analysisof the association rules between items of BRMand vessel acci-dents the importance of BRM are evaluated quantitativelyand the relationship of its items are minedThe contributionsof this paper are (1) proposing a cross level coding schemefor mining the multilevel association rules (2) presentingthe immune genetic algorithm using our cross level codingscheme for the research of BRM (3) finding the importantassociation rules of the items of BRM based on the basicmaritime investigation reports (4) providing suggestions toalleviate the incentive of shipwreck in aspect of seafarertraining assessment and management

The rest of this paper is organized into the followingsections In Section 2 our multilevel association rule miningbased on immune genetic algorithm is proposed In Section 3the proposed approach is simulated and analyzed then thefeasibility of the approach is illustrated by comparing withother methods some suggestions are developed toward themining results at the last part of this section In Section 4 weconclude our work

2 Multilevel Association Rule MiningBased on Immune Genetic Algorithm

In this paper we analyze items of BRM based on immunegenetic algorithm The goal is to mine the association rulesamong the items of BRM and the vessel accidents Howeverdue to the indirect data that can be collected the crosslevel coding scheme needs to be proposed In this section amultilevel association rule mining will be introduced wherethe basic data is considered as the bottom of the data miningand the summarized data that is items of BRM in thispaper is the middle level which has certain correspondingrelationship with the basic data

21 Multilevel Association Rule Mining Association rule isthe hiding relationship among the data items which is therelevance of different items appearance in the same event [1]Let 119868 = 119909

1 1199092 119909

119898 be a set of distinct literals called

items A set119883 sube 119868with 119896 = |119883| is called a 119896-item set or simplyan item set Let a database119863 be amultiset of subsets of 119868 Each119879 isin 119863 is called a transactionWe say that a transaction119879 isin 119863supports an item set 119883 sube 119868 if 119883 sube 119879 holds An associationrule is an expression119883 rArr 119884 where119883 and119884 are item sets and119883 cap 119884 = 0 holds The fraction of transactions 119879 supportingan item set119883with respect to database119863 is called the supportof X supp(119883) = |119879 isin 119863 | 119883 sube 119879||119863| The support of a rule119883 rArr 119884 is defined as

supp (119883 997904rArr 119884) = supp (119883 cup 119884) (1)

The confidence of this rule is defined as

conf (119883 997904rArr 119884) =supp (119883 cup 119884)supp (119883)

(2)

Abstract and Applied Analysis 3

Items X Class items CX

x1

x2

x3

x4

cx1

cx2

cx3

cxpxk

Figure 1 Mapping relationship of items and class items

In our work for a hierarchical organization to workeffectively and feasibility numbers of items in database aresummarized into class items as the higher level of theinformation which are interested by the user Then the items119868 = 119909

1 1199092 119909

119898 are mapped into the class items 119862119868 =

1198881199091 1198881199092 119888119909

119902 where 119902 is the number of class items with

119902 lt 119898 A set 119862119883 sube 119862119868 with 119901 = |119862119883| is called a 119901-class item set or simply a class item set Figure 1 shows usthe mapping relationship of items and class items Generallythere is more than one item corresponding to one class itemIn our paper binary coding scheme is used Each item will be0 or 1 in our scheme The link of items with different valuesforms a basic antibody The antibody is formed by the linkof class items with different values The value of class itemis decided by the product of the corresponding cross levelweights and the items which compose the class item Set119883 isa k-item set where each item is represented by 119909

119894 119909119894= 0 1

119862119883 is a p-class item set containing the class item 119888119909119895 The

contribution of 119909119894to 119888119909119895is 120572119894119895 called cross level weight where

sum119899

119895=1120572119894119895= 1 119899 is the total number of class items affected by

the item 119909119894 As shown in Figure 1 items 119909

1and 119909

3compose

the class item 1198881199091 the corresponding cross level weights are

12057211and12057231 while items 119909

1and 1199094compose the class item 119888119909

2

the corresponding cross level weights are 12057212and 120572

41 where

12057211+12057212= 1 and 120572

31= 12057241= 1 The value of 119888119909

119895is calculated

using the following formula

119888119909119895= [

[

sum119872

119894=1120572119894119895119909119894

(sum119870

119901=1120572119901119902119909119901)mas

times 119871]

]

(3)

where 119871 is the bit size of 119888119909119895 119872 is the number of items

who form the class item (sum119870119901=1120572119901119902119909119901)max

is the maximumsum value of the class item where 119870 is the number of itemscontained by the class item In formula (3) the symbol [sdot]represents the round calculation The code of 119888119909

119895can be

obtained by converting decimal of 119888119909119895into binary number

Definition 1 (information entropy) Because the immunesystem in the process of evolution is the uncertain onewhich is composed by antibodies the irregular degree ofthe immune system can be presented by average informationentropy of Shannons

Table 1 An example of basic antibody

Basic antibody 1199091

1199092

1199093

1199094

1199095

1199096

X(1) 1 1 0 0 0 1X(2) 0 1 1 0 1 1

Items Class itemsx1

x2

x3

x4

x5

x6

cx1

cx2

cx3

cx1 = 12057211x1 12057221x2 12057231x3

= 12057212x1 12057241x4 12057251x5

= 12057232x3 12057252x5 12057261x6

12057221 = 1 12057241 = 1 12057261 = 1

12057211 12057212= 08 02

12057231 12057232= 05 05

12057251 12057252= 03 07

cx2

cx3

Figure 2 An example of mapping relationship with 6 items and 3class items

Assuming an immune system is consisted with 119873 anti-bodies the number of gene for each antibody is 119872 The 119878numerals is used on encoding each gene where the binary isused in this paper that is 119878 = 2 and the 119894th code 119878

119894isin 0 1

where 119894 = 1 2 then the information entropy of these 119873antibodies is as follows

119867(119873) =1

119872

119872

sum119895=1

119867119895(119873) (4)

where119867119895(119873) = minussum

119878

119894=1119901119894119895log2119901119894119895 defined as the information

entropy of the jth gene 119901119894119895is the probably of the 119894th code

appeared in the jth geneIn order to explain the cross level coding scheme an

example of cross level coding from the basic antibodyto antibody is shown as follows The two complete basicantibodies are given in Table 1 As shown in Figure 2 wesummarized the 6 items into 3 class items the cross levelweights are also given When 119888119909

1= 120572111199091 120572211199092 120572311199093 and

12057211 12057221 12057231 = 08 1 05 the corresponding fragments

of the basic antibodies 119883(1) and 119883(2) are 08 1 0 and0 1 05 respectively Similarly the results can be derivedas 119888119909

1 1198881199092 1198881199093 = 18 02 1 for the antibody 119862119883(1) and

1198881199091 1198881199092 1198881199093 = 15 03 22 for the antibody 119862119883(2) Then

using the formula (3) where 119871 = 2119872 = 3 and 1198881199091has the

maximum sum value among all of the class items in 119862119883(1)then we have the equation

(

119870

sum119901=1

120572119901119902119909119901)

max

= 120572111199091+ 120572211199092+ 120572311199093= 18 (5)

where 119870 = 3 While 1198881199093has the maximum value of the same

one in 119862119883(2) then we have the equation

(

119870

sum119901=1

120572119901119902119909119901)

max

= 120572321199093+ 120572521199095+ 120572611199096= 22 (6)

Finally taking both values of (5) and (6) into the formula (3)the fragments of the antibodies 119862119883(1) and 119862119883(2) are coded


Table 2 The corresponding antibody of Table 1

Antibody 1198881199091

1198881199092

1198881199093

CX(1) 11 00 10CX(2) 11 00 11

as 1 1 and 1 1 which are shown in the second column ofTable 2The corresponding cross level codes of antibodies areshown in Table 2

After the cross level coding the mining of associationrules can be executed on the encoded population basedon the basic data However frequent patterns meeting thesupport requirements need to be found for association rulemining When there are multiply levels and dimensionslots of thresholds should be defined Meanwhile the searchspace storage space and running time will increase withthe increasing of the complexity of algorithm [11 12] Someintelligent algorithms such as genetic algorithm and artificialimmune algorithm can make the rule mining process moreefficient

22 Immune Genetic Algorithm for Association Rule MiningGenetic algorithm (GA) is a search heuristic that mimicsthe process of natural selection which generates solutionsto optimization problems using techniques inspired by nat-ural evolution such as inheritance mutation selection andcrossover In GA a population of candidate solutions to anoptimization problem is evolved towards better solutionsEach candidate solution has a set of properties which canbe mutated and altered The evolution usually starts froma population of randomly generated individuals and is aniterative process with the population in each iteration called ageneration In each generation the fitness of every individualin the population is evaluated and the fitness is usually thevalue of the objective function in the optimization problembeing solved However literature [13] notes the drawbacks ofGA such as the convergence direction is unable to control andthe object function does not correspond to constraints anddoes not have memory function

Artificial immune systems (AIS) are concerned withabstracting the structure and function of the immune systemto computational systems and investigating the applicationof these systems towards solving computational problemsfrommathematics engineering and information technology[14ndash17] Considering the characteristics of both GA andAIS many researchers have proposed the immune geneticalgorithm (IGA) with immune function [7 18] The aim ofleading immune concepts and methods into GA is theo-retically to utilize the locally characteristic information forseeking the ways and means of finding the optimal solutionwhen dealing with difficult problems To be exact it utilizesthe local information to intervene in the globally parallelprocess and restrain or avoid repetitive and useless workduring the course so as to overcome the blindness in actionof the crossover and mutation

Based on IGA the association rule mining algorithmfor BRM is researched in this paper We design a multilevelassociation rule mining algorithm especially for BRM using

IGA which is called IGA-BRM for simplicityThere are somedefinitions listed below about IGA-BRM algorithm

Definition 2 (antigen) Antigen is the molecule which can beidentified by any antibody or 119879 cell receptors correspondingto the problem Usually different problems have differentforms of data encoding In this paper it will be the attributevalues which we are interest in

Definition 3 (antibody) Antibody is the receptor whichis produced by B lymphocytes that react with antibodiesUsually the antibody corresponds to the optimal solution tothe problem In this paper it refers to association rules thatwill be mined

Definition 4 (affinity) The strength of the bind betweenantibodies is represented by affinity In AIS the more similarantibodies will have the bigger values of the affinity In orderto control the density of antibodies the new antibody iscreated only if the affinity between the two exceeds the affinitythreshold

The affinity between both antibodies can be calculatedusing (4) where 119873 = 2 Both antibodies are similar whenthe affinity between both antibodies is beyond the affinitythreshold 120582 which is called the similarity constant 09 le120582 le 1 The population similarity of antibodies also can beachieved using the same equation where119873 is the number ofantibodies in the population

Definition 5 (fitness) The strength of the bind between theantigen and the antibody which depends on how closely thetwo match is called fitness The more matching of the anti-body and the antigen is the stronger the molecular bindingwill be and the better the antibody can be recognized In IGAthe fitness indicates the adaptability of the candidate solutionsof objective function to the problem

Actually different fitness functions are chosen accordingto the different problemTherefore fitness function is the keyof convergence rate of the algorithm Assuming the supportof the antibody is Supp(119860) and the confidence is Conf(119860) thefitness of antibody 119860 is as follows

fitness (119860) = 119908119904Supp (119860) + 119908

119888Conf (119860) (7)

Definition 6 (antibody concentration) It is the proportion ofthe similar antibodies and the total number of antibodies inone generationWe use the symbol119862

119894to express the antibody

concentration of the 119894th antibodyWith the development of the IGA the antibodies in

the populations are more and more similar and the diver-sity of the antibodies is no longer maintaining the orig-inal level In order to maintain the diversity of the anti-bodies improve the global search ability and avoid theimmature convergence of the algorithm the fitness andconcentration of the generation should be controlled Theparameter called polymerization fitness is introduced tocontrol the selected probability of antibody in the geneticprocess


Definition 7 (polymerization fitness) It is the balance ofantibody concentration and the fitness of antibody which isexpressed by the following equation

fitness119901(119860119894) = fitness120583119862119894 (119860

119894) (8)

where 120583 is the correction factor of polymerization fitness and120583 le 0

Based on the above definitions and cross level encodingscheme steps of the IGB-BRM algorithm are described asfollows

Step 1 Initialize all the parameters such as the size ofpopulation119873 number of generation 119879 cross level weight 120572

119894119895

and affinity threshold 120582

Step 2 Process the data in database including encoding thebasic antibody and then converting to the antibody codeusing our cross level coding scheme

Step 3 Build the initial antibody population with half ran-domly generated and the other half taken from the coded datain Step 2

Step 4 Calculate the values of support confidence andfitness of each antibody using the coded data in Step 2

Step 5 Produce the new antibodies execute the GA algo-rithm which includes selection crossover and mutation

Step 6 Calculate the affinity of the new antibodies and thencompare them with the corresponding thresholds 120582

Step 7 If the comparing result of Step 6 is smaller than thethresholds it will execute Step 4 which means the antibodiessatisfying the requirement of diversity of population else itwill go to the next step

Step 8 Produce the other new antibodies which make theinformation entropy and the fitness satisfy the requiredthresholds

Step 9 Calculate the concentration and polymerization fit-ness of the antibody population and then select the 119873antibodies with the highest polymerization fitness as the newgeneration

Step 10 If the count number of generation 119905count ge 119879 theoptimal solution of antibodies will be achieved and then thealgorithm is over else 119905count adds 1 and goes to the Step 4 untilit is over

In the algorithm the antibodies with the highest poly-merization fitness will be selected as the new increasingantibodies when the antibody population is refreshed It isthe concentration based population refresh in IGA-BRMFrom formula (8) when the concentration is maintainedthe bigger the fitness is the more the polymerization fitnesswill be which will increase the selected probability of theantibody While the fitness is unchanged the polymerizationfitness will decrease with the increasing of the antibodyconcentration which will lead the selected probability to

be smaller This scheme will not only retain the antibodieswhich have excellent fitness but also inhibit the ones whoseconcentration is too high

3 Bridge Resource ManagementAnalysis Based on IGA-BRM

Resent years researches have shown that the composition ofmarine traffic is seaman ship and environment while thehuman error is the main factor which causes the accidentamong them In this paper based on vessel collision accidentson the sea we research the deep reasons induced by humanerrors The goal of our work is to reduce the maritimecollision accidents caused by human error in aspect ofseafarer training assessment and management

31 Model of IGA-BRM Traditionally the collision accidentsof vessels are analyzed by the data from the maritime investi-gation reports But the data does not reflect deep inducementswhich we want to guide seafarer training For examplea collision happened which was caused by the improperlookout when a duty officer is on dutywhile sailingMaritimesurvey results showed that the accident is mainly due to theimproper lookout which led to the vessel missing the primetime for the prevention of collision that is collision accidentcannot be avoided despite using the avoidance measures andgood seamanship when the vessel sailed into the distance toclosest approach (DCPA)Usually as themaritime competentauthority we want to make certain policy to reduce or evenavoid the vessel collision accidents through analyzing andsummarizing these reasons of accidents However in thesurface of the maritime accident investigation report theonly reason of the accident and the malpractice is classifiedas a human error accident which lacks the analysis of thehuman error itself In fact throughout the numerous collisionaccidents caused by the ldquoimproper lookoutrdquo there are manycauses hiding the ldquoimproper lookoutrdquoThe author who worksin seafarersrsquo competency examinationmanagement andmar-itime investigation for several years found that the mainfactors of ldquoimproper lookoutrdquo may be due to the fatigue poorawareness of risk situation caused by overconfidence and thebad work attitude and so on Based on the above reasonsIMO in STCW Manila amendments proposed concept ofBRM which summarized the causes of the accident into sixaspects called items of BRM in this paper listed in Table 3These items of BRM are only proposed and researchedqualitatively in STCWManila amendments In order tomakethe relationship between these items and the causes collectedin the daily maritime investigation clear we sorted data of thecauses into ten surface reasons of collision accident whichare listed in Table 4 formed data mining chain in this paperand then analyzed the data using the proposed IGA-BRMalgorithm quantificationally

32 Simulation Results and Analysis The data analyzed ischosen from the maritime investigation files in resent tenyears in north China There are about 1032 items Amongthem we randomly select 100 of ship collision accident


Table 3 Corresponding relationship between items in BRM andattribute

Attribute Items in BRM1198881199091

Team work abilities and working attitude1198881199092

Abilities of communication1198881199093

Decision-making capacity1198881199094

Pressure and fatigue1198881199095

Situational awareness and ability of risk assessment1198881199096

Educational background1198881199097

Severity of the accident

Table 4 Corresponding relationship between human errors andbasic attributes

Basic attribute Human error1199091

Improper lookout1199092

Nautical instruments use error1199093

Errors in judgment1199094

Inopportune navigating1199095

Improper communication1199096

Inappropriate maneuvering1199097

Not at safe speed1199098

Disobey navigation rules1199099

Unable to use good seamanship11990910

Error in positioning and route plan

investigation report and build a statistical table about thevalue of cross level weight that is120572

119894119895 as shown in Table 5The

ratio of cross level weight is counted by asking the maritimeinvestigation officer who takes part in the investigation of theaccident directly or calling to the crew in the accident report

In the simulation the items of BRM are consideredas attributes of the algorithm and the ten surface reasonscoming from the database are as the basic attributes whichhave certain contributions to the attributes Additionally tofind the relationship between items of BRM and the vesselcollision accidents the severity of the accident is added tothe attributes which can be achieved from the maritimeinvestigation reports denoted as 119888119909

7in Table 3 It is expressed

by the 2 bit binary code where 1198881199097= 11 represents the

worst accident We set the following size of population119873 =

50 correction factor of polymerization fitness 120583 = 08generation number 119879 = 100 affinity threshold 120582 = 01support threshold is 02 and confidence threshold is 08The IGA-BRM algorithm is executed ten times in Matlab65 Through the simulations about 27 association rulesare mined some interested rules with high support andconfidence are listed in Table 6

From the results in Table 6 the item of pressure andfatigue in BRM is the most important factor which leadsto the worst or most serious accident and team workabilities and working attitude are the second important factorthat causes the serious accident while the bad situationalawareness and ability of risk assessment are the third onecause of the general accident

Table 5 Values of 120572119894119895between human errors and items in BRM

1199091

1199092

1199093

1199094

1199095

1199096

1199097

1199098

1199099

11990910

1198881199091

02 02 0 01 0 02 07 0 06 041198881199092

02 0 01 01 04 0 0 07 0 01198881199093

01 02 02 02 0 02 0 0 0 01198881199094

04 05 03 04 05 03 01 01 02 051198881199095

01 0 04 02 0 03 02 02 01 01198881199096

0 01 0 0 01 0 0 0 01 01

In order to illustrate the reasonability of our method wecompare our results with the literature [1] which obtainedthe importance of BRM items by the expert questionnaireIn [1] the item of pressure and fatigue in BRM is the mostimportant factor in the management capacity evaluationsituational awareness and ability of risk assessment are thesecond one and team work abilities and working attitudeare the third one From both results whether in the vesselaccident analysis or management capacity evaluation theimportance of BRM items is basically consistent It meansthat our data mining method for BRM analysis is feasibleand reasonable We should pay more attention to the itemsof pressure and fatigue situational awareness and team workabilities of BRM in the maritime management

33 Suggestions Comparing with the actual situation thesimulation results in Table 6 are reasonable In terms of theitem of pressure and fatigue in BRM there are many factorsthat make the crew always in the state of pressure and fatiguesuch as the long-term work and boring life in the navigationthe worst natural conditions when sailing in the sea and thecontract period signedwith the shipping company of the crewbeing too long It is difficult to release the mental stress thatcauses every crew to have certain psychological subhealthmore or less Facing these problems many crews usuallyadopt the way of self-injury such as avoidance depressionand denial which make the crew always in the worst mentalstate In order to alleviate or solve these problems from theaspect of management we provide five suggestions as below

(1) Based on the Manila amendments the item aboutrest time of crew needs to be built the competentauthority should control the rest time longer than theminimum requirement of STCW strictly The PortState Control (PSC) should make the reasonable andfeasible ways to protect the rest and relax right of theseaman

(2) In terms of the vessel operators and owners consid-ering both the culture and character of the companythe control of the crewrsquos rest time and education ofmental health should be incorporated in the safetymanagement system in order to make the crew restenough and be happy

(3) According to the ldquopressure and fatiguerdquo there are twoways to solve the problem One is that IMO and thecompetent authority add the courses of mental healthto the standard of competency ability examination


Table 6 Association rules and the meaning of them based on BRM

Association rule withsupport and confidence Meaning based on BRM

1198881199091= 11 rArr 119888119909

7= 10

[supp = 652 conf = 881] The worst team work abilities and working attitude lead to the serious accident

1198881199094= 11 rArr 119888119909

7= 11

[supp = 693 conf = 890] The worst pressure and fatigue lead to the worst accident

1198881199095= 10 rArr 119888119909

7= 01

[supp = 558 conf = 863]The bad situational awareness and ability of risk assessment lead to the generalaccident

1198881199091= 11 cap 119888119909

5= 10 rArr 119888119909

7= 11

[supp = 326 conf = 821]The worst team work abilities and attitude added to bad situational awareness andability of risk assessment lead to the worst accident

1198881199093= 10 cap 119888119909

4= 10 rArr 119888119909

7= 10

[supp = 218 conf = 854]The bad pressure and fatigue added to bad decision-making capacity lead to theserious accident

1198881199092= 10 cap 119888119909

6= 01 rArr 119888119909

7= 01

[supp = 252 conf = 819]The bad abilities of communication and poor educational background lead to thegeneral accident

1198881199094= 11 cap 119888119909

5= 10 rArr 119888119909

7= 10

[supp = 336 conf = 819]The worst pressure and fatigue added to situational awareness and ability of riskassessment lead to the serious accident

for the crew and the other is to modify the item oftheminimumsafety requirements for vessels forcedlyrequire the vessel of which voyage exceeds a certaintime and provide the mental health counselor

(4) The shipping company or the crew managementcompany should authorize the master to coordinatethe sailing team in order to improve the ability ofteamwork and enhance the obedience consciousnessof the member Coordinate content should focuson breakthrough teamworkrsquos five obstacles namedldquolack of trustrdquo ldquofear of conflictrdquo ldquolack of investmentrdquoldquoescape responsibilityrdquo ldquoignore resultrdquoThe competentauthority should evaluate the leadership organiza-tion and coordination capacity quantitatively andbring the ability into the master competency abilityrequirement

(5) In order to enhance the crewrsquos ability of dealing withcrisis maritime education should focus on the situa-tional awareness Navigation is the dynamic processand there are so many emergencies which demandthe crew to develop the excellent response abilityTherefore it is necessary to add the operation ofnavigation simulator to the maritime education andincrease the investment of the study of navigationsimulator for the better simulation result

4 Conclusion

In this paper based on immune genetic algorithm wepropose a multilevel association rule mining for BRMannounced by IMO in 2010 named IGA-BRM The goal ofthis paper is to mine the association rules among the items ofBRM and the vessel accidents

Because of the indirect data that can be collected whichseems useless for the analysis of the relationship betweenitems of BIM and the vessel accidents a cross level codingscheme for mining the multilevel association rules is pro-posed By this way based on basic maritime investigation

data the relationship between items of BRM and vesselaccidents is built We execute the IGA-BRM with the crosslevel coding scheme for the research of BRM As a resultbased on the basic maritime investigation reports manyimportant association rules about the items of BRM aremined Among these association rules we find that theitem of pressure and fatigue in BRM is the most importantfactor which leads to the worst or most serious accidentand team work abilities and working attitude are the secondimportant factor that causes the serious accident while thebad situational awareness and ability of risk assessment arethe third one that causes the general accident and so on Atlast of the paper according to the pattern that mined usingIGA-BRM algorithm five suggestions are provided for thework of seafarer training assessment and management onthe maritime competent authority level

Conflict of Interests

The authors declare that there is no conflict of interestsregarding the publication of this paper

References

[1] Y L Liu ldquoRating model of resource management competencyof seafarersrdquoNavigation of China vol 36 no 2 pp 82ndash85 2013

[2] B Liu W Hsu and Y Ma ldquoIntegrating classification andassociation rule miningrdquo in Proceedings of the ACM KnowledgeDiscovery and Data Mining (KDD rsquo98) pp 80ndash86 1998

[3] B Liu M Hu andW Hsu ldquoIntuitive representation of decisiontrees using general rules and exceptionsrdquo in Proceedings of the7th National Conference on Artificial Intellgience pp 615ndash620July 2000

[4] J Hipp U Guntezr and G Nakhaeizadeh ldquoAlgorithms forassociation rule miningmdasha general survey and comparisonrdquoin Proceedings of the 6th ACM International Conference onKnowledge Discovery and Data Mining Boston Mass USA2000

[5] S Yin G Wang and H Karimi ldquoData-driven design of robustfault detection system for wind turbinesrdquoMechatronics 2013


[6] R Rastogi and K Shim ldquoMining optimized association ruleswith categorical and numeric attributesrdquo IEEE Transactions onKnowledge and Data Engineering vol 14 no 1 pp 29ndash50 2002

[7] HMo and L Xu ldquoImmune clone algorithm formining associa-tion rules on dynamic databasesrdquo inProceedings of the 17th IEEEInternational Conference on Tools with Artificial Intelligence(ICTAI rsquo05) pp 202ndash206 Hong Kong China November 2005

[8] T D Do S C Hui A C M Fong and B Fong ldquoAssociativeclassification with artificial immune systemrdquo IEEE Transactionson Evolutionary Computation vol 13 no 2 pp 217ndash228 2009

[9] B Liu M Hu and W Hsu ldquoMulti-level organization andsummarization of the discovered rulesrdquo in Proceedings of the6th ACM International Conference on Knowledge Discovery andData Mining (KDD rsquo01) pp 91ndash100 Boston Mass USA August2000

[10] S Yin S X Ding A H A Sari and H Hao ldquoData-drivenmonitoring for stochastic systems and its application on batchprocessrdquo International Journal of Systems Science vol 44 no 7pp 1366ndash1376 2013

[11] S Yin H Luo and S X Ding ldquoReal-time implementation offault-tolerant control systems with performance optimizationrdquoIEEE Transactions on Industrial Electronics vol 64 no 5 pp2402ndash2411 2014

[12] S Yin S Ding A Haghani H Hao and P Zhang ldquoA com-parison study of basic datadriven fault diagnosis and processmonitoring methods on the benchmark Tennessee Eastmanprocessrdquo Journal of Process Control vol 22 no 9 pp 1567ndash15812012

[13] P Hajela J Yoo and J Lee ldquoGA based simulation of immunenetworks applications in structural optimizationrdquo EngineeringOptimization vol 29 no 1ndash4 pp 131ndash149 1997

[14] LWang J Pan and L Jiao ldquoImmune algorithmrdquoActa Electron-ica Sinica vol 28 no 7 pp 74ndash78 2000

[15] J Timmis M Neal and J Hunt ldquoAn artificial immune systemfor data analysisrdquo BioSystems vol 55 no 1ndash3 pp 143ndash150 2000

[16] S M Garrett ldquoHow do we evaluate artificial immune systemsrdquoEvolutionary Computation vol 13 no 2 pp 145ndash178 2005

[17] V Cutello G Nicosia M Pavone and J Timmis ldquoAn immunealgorithm for protein structure prediction on lattice modelsrdquoIEEE Transactions on Evolutionary Computation vol 11 no 1pp 101ndash117 2007

[18] L Jiao and L Wang ldquoA novel genetic algorithm based onimmunityrdquo IEEE Transactions on SystemsMan and CyberneticsA vol 30 no 5 pp 552ndash561 2000

Submit your manuscripts athttpwwwhindawicom

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

MathematicsJournal of


Mathematical Problems in Engineering

Hindawi Publishing Corporationhttpwwwhindawicom

Differential EquationsInternational Journal of

Volume 2014

Applied MathematicsJournal of


Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of


Mathematical PhysicsAdvances in

Complex AnalysisJournal of


OptimizationJournal of


CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

International Journal of


Operations ResearchAdvances in

Journal of


Function Spaces

Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

International Journal of Mathematics and Mathematical Sciences


The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014


Algebra

Discrete Dynamics in Nature and Society



Decision SciencesAdvances in

Discrete MathematicsJournal of


Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Stochastic AnalysisInternational Journal of


classification and associationmining techniques according tothe target of discovery are not predetermined in associationrule mining The integration is done by focusing on mininga special subset of association rules called class associationrules (CARs) An efficient algorithm is also given for buildinga classifier based on the set of discovered CARs In [3]an efficient representation scheme called general rules andexceptions is proposed This paper focused on using therepresentation to express the knowledge embedded in adecision tree In this representation a unit of knowledgeconsists of a single general rule and a set of exceptions Thisscheme reduces the complexity of the discovered knowledgesubstantially It is also intuitive and easy to understandAuthors in literatures [4 5] deal with the algorithmic aspectsof association rulemining and they explain the fundamentalsof association rule mining and moreover derive approachesPaper [6] presents effective techniques for pruning the searchspace when computing optimized association rules for bothcategorical and numeric attributes However to mine theassociation rules effectively it needs to tackle the problemson the very large search space of candidate rules during therule discovery process Some intelligent searching methodsare used in the rule discovery process such as literatures[7 8] The paper [7] seeks to generate large item sets ina dynamic transaction database using immune clone algo-rithm where intratransactions intertransactions and dis-tributed transactions are considered for mining associationrules The paper [8] proposes an approach which is basedon artificial immune systems for mining association ruleseffectively for classification Instead of massively searchingfor all possible association rules this approach only findsa subset of association rules that are suitable for systemeffective in an evolutionary manner However producing toomany rules have been regarded as a major problem withmany data mining techniques The large number of rulesmakes it very hard for manual inspection of the rules togain insight of the domain Many data mining methodsmentioned above typically generate a list of rules with nofurther organization It is like writing a book or a paper byrandomly listing out all the facts and formulas The resultingbook will be difficult for anyone to read because one cannotsee any relationships among the facts and formulas In papers[9 10] they proposed a technique to intuitively organize andsummarize the discovered rules With this organization thediscovered rules can be presented to the user in the way asthe thinking and talking about knowledge in the daily livesA common strategy for organizing the large number of rulesis by hierarchy In such an organization the rules are summa-rized into topics and presented at different levels of detailsThe main advantage of the hierarchical organization is that itallows the user to view the problem from general to specificand to focus hisher attention on those interesting aspectsPapers [9 10] manage the complexity of the discovered rulesin a hierarchical fashion but these methods are too complexto deal with the maritime accident data In our paper weuse the hierarchical thinking organizing lots of items intomultiple levels by the cross level coding Additionally theimmune genetic algorithm is used in the rulemining processBy this way the major association rules which reflect deep

inducements of accidents will be mined effectively in anevolutionary manner

This paper mainly researches BRMThrough the analysisof the association rules between items of BRMand vessel acci-dents the importance of BRM are evaluated quantitativelyand the relationship of its items are minedThe contributionsof this paper are (1) proposing a cross level coding schemefor mining the multilevel association rules (2) presentingthe immune genetic algorithm using our cross level codingscheme for the research of BRM (3) finding the importantassociation rules of the items of BRM based on the basicmaritime investigation reports (4) providing suggestions toalleviate the incentive of shipwreck in aspect of seafarertraining assessment and management

The rest of this paper is organized into the followingsections In Section 2 our multilevel association rule miningbased on immune genetic algorithm is proposed In Section 3the proposed approach is simulated and analyzed then thefeasibility of the approach is illustrated by comparing withother methods some suggestions are developed toward themining results at the last part of this section In Section 4 weconclude our work

2 Multilevel Association Rule MiningBased on Immune Genetic Algorithm

In this paper we analyze items of BRM based on immunegenetic algorithm The goal is to mine the association rulesamong the items of BRM and the vessel accidents Howeverdue to the indirect data that can be collected the crosslevel coding scheme needs to be proposed In this section amultilevel association rule mining will be introduced wherethe basic data is considered as the bottom of the data miningand the summarized data that is items of BRM in thispaper is the middle level which has certain correspondingrelationship with the basic data

21 Multilevel Association Rule Mining Association rule isthe hiding relationship among the data items which is therelevance of different items appearance in the same event [1]Let 119868 = 119909

1 1199092 119909

119898 be a set of distinct literals called

items A set119883 sube 119868with 119896 = |119883| is called a 119896-item set or simplyan item set Let a database119863 be amultiset of subsets of 119868 Each119879 isin 119863 is called a transactionWe say that a transaction119879 isin 119863supports an item set 119883 sube 119868 if 119883 sube 119879 holds An associationrule is an expression119883 rArr 119884 where119883 and119884 are item sets and119883 cap 119884 = 0 holds The fraction of transactions 119879 supportingan item set119883with respect to database119863 is called the supportof X supp(119883) = |119879 isin 119863 | 119883 sube 119879||119863| The support of a rule119883 rArr 119884 is defined as

supp (119883 997904rArr 119884) = supp (119883 cup 119884) (1)

The confidence of this rule is defined as

conf (119883 997904rArr 119884) =supp (119883 cup 119884)supp (119883)

(2)



x1

x2

x3

x4

cx1

cx2

cx3

cxpxk



1 1199092 119909


1198881199091 1198881199092 119888119909



119894 119909119894= 0 1



sum119899



1and 119909

3compose


12057211and12057231 while items 119909


2


41 where

12057211+12057212= 1 and 120572

31= 12057241= 1 The value of 119888119909

119895is calculated


119888119909119895= [

[

sum119872

119894=1120572119894119895119909119894

(sum119870

119901=1120572119901119902119909119901)mas

times 119871]

]

(3)




119895can be





1199092

1199093

1199094

1199095

1199096

X(1) 1 1 0 0 0 1X(2) 0 1 1 0 1 1

Items Class itemsx1

x2

x3

x4

x5

x6

cx1

cx2

cx3

cx1 = 12057211x1 12057221x2 12057231x3

= 12057212x1 12057241x4 12057251x5

= 12057232x3 12057252x5 12057261x6

12057221 = 1 12057241 = 1 12057261 = 1

12057211 12057212= 08 02

12057231 12057232= 05 05

12057251 12057252= 03 07

cx2

cx3



119894isin 0 1


119867(119873) =1

119872

119872

sum119895=1

119867119895(119873) (4)

where119867119895(119873) = minussum

119878





1= 120572111199091 120572211199092 120572311199093 and







(

119870

sum119901=1

120572119901119902119909119901)

max

= 120572111199091+ 120572211199092+ 120572311199093= 18 (5)



(

119870

sum119901=1

120572119901119902119909119901)

max

= 120572321199093+ 120572521199095+ 120572611199096= 22 (6)




Antibody 1198881199091

1198881199092

1198881199093

CX(1) 11 00 10CX(2) 11 00 11













fitness (119860) = 119908119904Supp (119860) + 119908

119888Conf (119860) (7)







fitness119901(119860119894) = fitness120583119862119894 (119860

119894) (8)




119894119895

















































1199091

1199092

1199093

1199094

1199095

1199096

1199097

1199098

1199099

11990910

1198881199091

02 02 0 01 0 02 07 0 06 041198881199092

02 0 01 01 04 0 0 07 0 01198881199093

01 02 02 02 0 02 0 0 0 01198881199094

04 05 03 04 05 03 01 01 02 051198881199095

01 0 04 02 0 03 02 02 01 01198881199096

0 01 0 0 01 0 0 0 01 01









1198881199091= 11 rArr 119888119909

7= 10


1198881199094= 11 rArr 119888119909

7= 11


1198881199095= 10 rArr 119888119909

7= 01


1198881199091= 11 cap 119888119909

5= 10 rArr 119888119909

7= 11


1198881199093= 10 cap 119888119909

4= 10 rArr 119888119909

7= 10


1198881199092= 10 cap 119888119909

6= 01 rArr 119888119909

7= 01


1198881199094= 11 cap 119888119909

5= 10 rArr 119888119909

7= 10





4 Conclusion






References



























Volume 2014




Journal of











Journal of


Function Spaces






Algebra











x1

x2

x3

x4

cx1

cx2

cx3

cxpxk



1 1199092 119909


1198881199091 1198881199092 119888119909



119894 119909119894= 0 1



sum119899



1and 119909

3compose


12057211and12057231 while items 119909


2


41 where

12057211+12057212= 1 and 120572

31= 12057241= 1 The value of 119888119909

119895is calculated


119888119909119895= [

[

sum119872

119894=1120572119894119895119909119894

(sum119870

119901=1120572119901119902119909119901)mas

times 119871]

]

(3)




119895can be





1199092

1199093

1199094

1199095

1199096

X(1) 1 1 0 0 0 1X(2) 0 1 1 0 1 1

Items Class itemsx1

x2

x3

x4

x5

x6

cx1

cx2

cx3

cx1 = 12057211x1 12057221x2 12057231x3

= 12057212x1 12057241x4 12057251x5

= 12057232x3 12057252x5 12057261x6

12057221 = 1 12057241 = 1 12057261 = 1

12057211 12057212= 08 02

12057231 12057232= 05 05

12057251 12057252= 03 07

cx2

cx3



119894isin 0 1


119867(119873) =1

119872

119872

sum119895=1

119867119895(119873) (4)

where119867119895(119873) = minussum

119878





1= 120572111199091 120572211199092 120572311199093 and







(

119870

sum119901=1

120572119901119902119909119901)

max

= 120572111199091+ 120572211199092+ 120572311199093= 18 (5)



(

119870

sum119901=1

120572119901119902119909119901)

max

= 120572321199093+ 120572521199095+ 120572611199096= 22 (6)




Antibody 1198881199091

1198881199092

1198881199093

CX(1) 11 00 10CX(2) 11 00 11













fitness (119860) = 119908119904Supp (119860) + 119908

119888Conf (119860) (7)







fitness119901(119860119894) = fitness120583119862119894 (119860

119894) (8)




119894119895

















































1199091

1199092

1199093

1199094

1199095

1199096

1199097

1199098

1199099

11990910

1198881199091

02 02 0 01 0 02 07 0 06 041198881199092

02 0 01 01 04 0 0 07 0 01198881199093

01 02 02 02 0 02 0 0 0 01198881199094

04 05 03 04 05 03 01 01 02 051198881199095

01 0 04 02 0 03 02 02 01 01198881199096

0 01 0 0 01 0 0 0 01 01









1198881199091= 11 rArr 119888119909

7= 10


1198881199094= 11 rArr 119888119909

7= 11


1198881199095= 10 rArr 119888119909

7= 01


1198881199091= 11 cap 119888119909

5= 10 rArr 119888119909

7= 11


1198881199093= 10 cap 119888119909

4= 10 rArr 119888119909

7= 10


1198881199092= 10 cap 119888119909

6= 01 rArr 119888119909

7= 01


1198881199094= 11 cap 119888119909

5= 10 rArr 119888119909

7= 10





4 Conclusion






References



























Volume 2014




Journal of











Journal of


Function Spaces






Algebra











Antibody 1198881199091

1198881199092

1198881199093

CX(1) 11 00 10CX(2) 11 00 11













fitness (119860) = 119908119904Supp (119860) + 119908

119888Conf (119860) (7)







fitness119901(119860119894) = fitness120583119862119894 (119860

119894) (8)




119894119895

















































1199091

1199092

1199093

1199094

1199095

1199096

1199097

1199098

1199099

11990910

1198881199091

02 02 0 01 0 02 07 0 06 041198881199092

02 0 01 01 04 0 0 07 0 01198881199093

01 02 02 02 0 02 0 0 0 01198881199094

04 05 03 04 05 03 01 01 02 051198881199095

01 0 04 02 0 03 02 02 01 01198881199096

0 01 0 0 01 0 0 0 01 01









1198881199091= 11 rArr 119888119909

7= 10


1198881199094= 11 rArr 119888119909

7= 11


1198881199095= 10 rArr 119888119909

7= 01


1198881199091= 11 cap 119888119909

5= 10 rArr 119888119909

7= 11


1198881199093= 10 cap 119888119909

4= 10 rArr 119888119909

7= 10


1198881199092= 10 cap 119888119909

6= 01 rArr 119888119909

7= 01


1198881199094= 11 cap 119888119909

5= 10 rArr 119888119909

7= 10





4 Conclusion






References



























Volume 2014




Journal of











Journal of


Function Spaces






Algebra











fitness119901(119860119894) = fitness120583119862119894 (119860

119894) (8)




119894119895

















































1199091

1199092

1199093

1199094

1199095

1199096

1199097

1199098

1199099

11990910

1198881199091

02 02 0 01 0 02 07 0 06 041198881199092

02 0 01 01 04 0 0 07 0 01198881199093

01 02 02 02 0 02 0 0 0 01198881199094

04 05 03 04 05 03 01 01 02 051198881199095

01 0 04 02 0 03 02 02 01 01198881199096

0 01 0 0 01 0 0 0 01 01









1198881199091= 11 rArr 119888119909

7= 10


1198881199094= 11 rArr 119888119909

7= 11


1198881199095= 10 rArr 119888119909

7= 01


1198881199091= 11 cap 119888119909

5= 10 rArr 119888119909

7= 11


1198881199093= 10 cap 119888119909

4= 10 rArr 119888119909

7= 10


1198881199092= 10 cap 119888119909

6= 01 rArr 119888119909

7= 01


1198881199094= 11 cap 119888119909

5= 10 rArr 119888119909

7= 10





4 Conclusion






References



























Volume 2014




Journal of











Journal of


Function Spaces






Algebra









































1199091

1199092

1199093

1199094

1199095

1199096

1199097

1199098

1199099

11990910

1198881199091

02 02 0 01 0 02 07 0 06 041198881199092

02 0 01 01 04 0 0 07 0 01198881199093

01 02 02 02 0 02 0 0 0 01198881199094

04 05 03 04 05 03 01 01 02 051198881199095

01 0 04 02 0 03 02 02 01 01198881199096

0 01 0 0 01 0 0 0 01 01









1198881199091= 11 rArr 119888119909

7= 10


1198881199094= 11 rArr 119888119909

7= 11


1198881199095= 10 rArr 119888119909

7= 01


1198881199091= 11 cap 119888119909

5= 10 rArr 119888119909

7= 11


1198881199093= 10 cap 119888119909

4= 10 rArr 119888119909

7= 10


1198881199092= 10 cap 119888119909

6= 01 rArr 119888119909

7= 01


1198881199094= 11 cap 119888119909

5= 10 rArr 119888119909

7= 10





4 Conclusion






References



























Volume 2014




Journal of











Journal of


Function Spaces






Algebra












1198881199091= 11 rArr 119888119909

7= 10


1198881199094= 11 rArr 119888119909

7= 11


1198881199095= 10 rArr 119888119909

7= 01


1198881199091= 11 cap 119888119909

5= 10 rArr 119888119909

7= 11


1198881199093= 10 cap 119888119909

4= 10 rArr 119888119909

7= 10


1198881199092= 10 cap 119888119909

6= 01 rArr 119888119909

7= 01


1198881199094= 11 cap 119888119909

5= 10 rArr 119888119909

7= 10





4 Conclusion






References



























Volume 2014




Journal of











Journal of


Function Spaces






Algebra






























Volume 2014




Journal of











Journal of


Function Spaces






Algebra
















Volume 2014




Journal of











Journal of


Function Spaces






Algebra









Documents

Research Article Multilevel Association Rule Mining for Bridge …downloads.hindawi.com/journals/aaa/2014/278694.pdf · 2019-07-31 · Research Article Multilevel Association Rule