Application Association Rule Mining in Medical- Ьiological Investigations: а Survey

Abstract

1n thischapterа, survey isproposedrelatedto eth applicatioonf Apriori -likealgorithms inmedicalnd biological vinestigatiorfons discovering erfquent setsof atributes valuesindata and thenextractinglogicarules in eth form of implicatidoenpendencibeestween valuesof observed andmeasured atbrtiutes dan diaognstic pmareter Intelligentdata processingisnow an integralpart of ЬiomedicarlesearchR.evealing diferent dependencieisnthe data (implicative, nfuctional, coпelational,etc.)helpsin diagnosistre,atment's planning, predictintghe course of diseasesand inidentifying new factors that expand the understanding of specialistasbout specicfidiseasesand theircomЬinations. The purpose of many Ьiomedicasl tudiesisto higblightassociativreules inа given data set. The associatiornule isthe rule inthe form Х ⇒ У, whereХ and У are nonintersectinsgetsof distinclitteralscalled items.1n general case,we canconsiderа set of itemsas а set of all aitrbutes' values that can appear indescriptionosf some objectsor situations(atrnsactions) inа data base. Let 1= {i1, Ь, ... , iN}Ье а set of items.А database(TDB) isа setof transactions, hwere rtansaction <tid,Х> containsа setof items(i.e.,Х 1) and isassociatedwith а uniqueidentifiertid.А non-emptyitemseУt 1 isа q-itemseitfitcontainsq items. А transaction <tid,Х> issaidto containitemseУt ifУ Х. The number of transactionsinTDB containingitemsetХ iscalled the support of Х, denoted as sup(X): sup (Х)= l{tidl(tid,У) Е TDB, Х Y}I, wherelsl denotesthe cardinalityof s. Giving а minimumsupport threshold, min-sup, an itemУ isfrequent ifsup(Y) 2: min-sup. In frequent itemsetsa,ssociatiornules are extracted inthe form of implicationsrfo, whichthe value of support isan iomprttan chaarcteristic:Sup(rule)= sup(X ⇒ У)= sup(X u У). The rule has а measureof reliaЬilitycalled confidence and defined as ofllows: conf(X ⇒ У)= sup(X u Y)/sup(X). Condfience isdefined as the part of all rtansactions inTDB containingХ and У, among thosetransactionsthat containХ.

Keywords

• Apriori algorithm

-

Introduction

The third main meitrc of association rulesis Lift defined as fo llows:lift(X ⇒ У) = conXf( ⇒ Y)/sup(Y). This measure quantifies the predictive power of ruleХ ⇒ У.

The traditionalpurpose of data analysisis to find allassociative rules(ASRs) that have support and confidence above the specified minimum values.

In this paper, we consider the applicationsof the classicalrApiori algorithm for mining ASRs in Ьiomedical data. The choice of this algorithmis justified Ьу the fol­ lowingarguments: it is the most understandaЫe and easy mastered Ьу specialists in diefrent applied efilds, it is iunversal, in the sense that ASRs containing in frequent itemsets сап Ье extracted with the helpofthis algorithm [ 30 ]; it is constantly improv­ ing based on many progressive techniques in ASR mining.

The idea ofthe classicalApriori algorithm[ 2 ] is based on the following considera­ tion: q-itemset сап Ье frequent if and onlyif allits proper sub-itemsets are frequent.

At the first step ofthe algorith щ allthe items are considered (valuesof atributes, elementsoftransactions) and among these are separated those satisfying the condition of minimal support. Then the separated items ear used to fo rm itemsets of two items (candidates fo r frequency). For theщ the support is calculatedand those that do not meet the minimum support are removed. The remaining itemsets are used to form ones of three items. The process is going on iteratively, as long as it is possiЫe to generate а new set of candidates rfo rfequency. The Apriori algorithmuses an induc­ tive method ofconstructing sets ofthe cardinality (q+l) ((q+l)-sets) from their sub­ sets of the cardinalityq (q-sets). The method of rfoming (q +1)-sets from q-sets and calculatingtheir supports are the main sub-processes of the rApiori algorithm deter­ mining its computationalcomplexity.

The paper is organized as follows. The Section 2, 3 contain brief surveys of apply­ ing the Apriori algorithmin medical studies. The perfection of the Apriori algorithm is considered in Section 4 followed Ьу а smallconclusionsection. 2

The Apriori Algorithm in Medical Studies The Apriori algorithmis widelyused in medicalresearch. These studies cover: cardi­ ovascular disease [ 23, 29, 41 ]; lungcancer [ 18 ]; oralcancer [ 39 ]; infectious diseases (Ebolavirus) [ 13 ]; type 2 diabetes [ 26, 34, 37, 45, 46 ]; Alzheimer's disease [ 7 ]; liver cancer [ 32 ]. In [ 41 ], some other diseases are enumerated for the study ofwhich were used the Apriori algorithm or its modicfiations: asthma, impotence, lupus, obesity, whooping cough, pregnancy, phenomenon Raynaud syndrome. The proЫems solved are also varied: searching fo r unknown ertnds in disease; determining the nature of disease based on а prediction method; diagnosis (detection) of disease; predicting а patient's response to drug [ 38, 44 ]; elary diagnosis and prevention of disease [ 37 ]; prediction of illness'sprogress (course of disease); predicting the outcome of disease [ 23 ]; identicfiation of disease risk factors [ 34 ]; identification ofrelationships between diferent medical operations, appointments, analyses and diagnoses [ 35 ]; extracting diagnostic pattems (sets of fe atures, symptoms) and association rules in electronic medicaldatabase [ 1, 11, 19, 22 ] and many others.

The papers [ 3, 5, 41, 51 ] give detailed reviews of ASR mining based on Apriori al­ gorithrn and its modicfiations. In [ 39 ], early detection and prevention of oral cancer is considered and the Apriori algorithrn is used to assess the chcane of patients' survival. This is achieved Ьу extracting а set of significant rules among various laboratory tests and investigations like FNAC (Fine Needle Aspiration Cytology) of neck node, LFT (Liver Function Tests), Biopsy, USG (Ultra-SonoGraphy), СТ scan or MRI (computer tomography), and survivaЬility of the oral cancer patients. Liver Function Tests (LFT) give information about the state of а patient's liver. Most liver diseases cause only mild symptoms initially, but it is vital that these diseases Ье detected early. Biopsy is important, as it is the only sure way to know if the abnormal area is cancer. USG is an ultrasound based diagnostic imaging technique to visualize intra-abdominal sctru­ tures. The extracted ASRs clearly show that if FNAC of neck node, USG and СТ scan/ МR1 is positive, then chance of survival is reduced. However, if LFT is normal, the probaЬility of survival is high. All the generated rules hold the highest confidence level.

Nahar J. et al. [ 31 ] extract the significant prevention factors for particular types of cancer: Ыadder, breast, cervical, lung, prostate and skin cancer. The algorithms Apri­ ori, Predictive Apriori, and Tertius are used to discover most of the significant pre­ vention factors against these specific types of cancer. Predictive rApiori tries to max­ imize expected accaurcy rather than confidence in Apriori, Tertius is а top-down rule discovery system employing well known decision tree algorit.hrn

The article [ 6 ] describes the methods to detect the comorЬidity of professional dis­ eases caused Ьу pathogenic factors, namely, Ьу ionizing radiation. rApiori algoritshrn implemented in the environment of the SQL Server Analysis Services Data Mining. The following deceases are considered: chronic radiation sickness of deegre 1, 2 and 3, residual phenomena of chronic radiation sickness, exposure to ionizing radiation, malaise and aftigue, and vegetative nervous system disorders.

uNmerous papers deal with the use of the Apriori algorithrn to study coronary heart disease. Karaolis et al. [ 20 ] developed а data analysis system to search for associa­ tions to assess the heart disease risk cfators with WЕКА tools. In [ 29 ], the objective of the study was to efectively predict possiЫe heart attacks from а patient dataset. The Data Base consisted of 209 records (instances) collected ofrm а hospital in Iran and 8 attributes. Apriori algorithrns were implemented in WЕКА 2016 (version 3.9.0) and МАТLАВ R2013a software. The preprocessing of data consisted of previous purifying data through the Discretization Unsupervised Filter and а discretizing meth­ od to chgane numeric data into nominal data. The algorithrn implemented in МАТLАВ showed the best results related to the diagnosis prediction accuracy Ьу the use of the obtained ASRs.

It should Ье noted that there are still very few works in the domestic literature on the use of the Apriori algoritm in Ьiomedical research. 1n addition to the work of Bi­ ryukov А. and Dumansky S. [ 6 ], we сап cite the article [ 5 ], which proposes а new efective algorithrn AprioriScale to build ASRs. The algorithrn is applied to the prob­ lem of detecting children's diseases: obesity and metabolic syndrome. Two proЫems are solved: 1) the possiЬility of distinguishing between obesity and metabolic syn­ drome, and 2) identifying а comЬination of risk factors determined in the early stages of а child's development that may indicate the onset of the disease in adolescence. 12 rules with the reliaЬility above 0.75 were obtained. Let's take а closer look at two of these rules: • •

Toxicosis, ПР, Weak:LD➔ MS frequency = 0,16, probaЬility = 0,83; Toxicosis, ПР, ExtraGM, РРН ➔ MS frequency = 0,09, probaЬility = 0,93, where ПР - threat of interuption of pregnancy, MS - metabolic syndrome, WeakLD - weak labour delivery, and ExtraGM - extragenital diseases of mother, РРН is post­ prandial hyperlipernia. These rules show that if factors such as toxicosis, ПР, Ex­ rtaGM, later accompanied Ьу factors such as Weak:LD, РРН are observed in the elary stages of а child's development, the risk of metabolic sdynrome is high, the probaЬil­ ity of its nrnaifestation is between 0.83 and 0.93. 3

Analysis of Biological and Genetic Data Based on Association uRle Extracting А topic dealing with the analysis of patient Ьiological data is now becorning particu­ larly relevant. The Ьiological data analysis is connected with the identification of previously unkown hidden patterns (frequent itesrnets), associative structures in the large number of Ьiological sequences. These sequences include gene sequences, arni­ no acid seequnces, protein composition, and other data that display the structure, lo­ calization, interaction or functioning of proteins and genes in cells. Arnino acids are the building material of proteins. The shape and other properties of proteins are asso­ ciated with the exact sequence of anrnio acids contained in them. The chernical prop­ erties of amino acids deternine the Ьiological activity of proteins.

Many diseases have Ьiological nature: obesity, high Ыооd cholesterol, diabetes, in­ soimna, hartritis, nad many others. Analysis of gene information, including Apriori algorithms, [ 1, 10, 21, 27 ] helps to study the nature of disease, optizrnie its treatment, predict the course of disease. An overview of some methods of extracting knowledge rfom Ьiological (DNA) sequences is given in [ 8 ]. The comparison of the rApiori algo­ rithm with other algorithms in the mutation analysis is produced in [ 28 ]. In [ 17 ], а model is proposed rfo nifding а dorninant sequence of arnino acids to Ыосk the growth of cancer cells based on protein clustering.

In [ 16 ], а comparative analysis of classifiers in cancer prediction using multiple da­ ta rnining techniques is given. The dataset contained а total of 844 records and nine efatures: Clump thickness, Unifornity of cell size, Unifornity of cell shape, Marginal adhesion, Single epithelial cell size, Ваге nuclei, Вland chrornatin, Nornal nucleoli and Mitoses. The analysis of data consisted of two stages. The rfist stage applies the algorithm rApiori to reduce the number of iunpt efatures. In the second stage, six classifiers have been applied and validated through а k-fold cross-validation scheme. Some variaЫes named as Marginal adhesion and Ваге nuclei have been removed as the noise data. Thus, а new subset of seven features has been provided rfo solving the proЫem of classification. Six predictive algorithms were chosen: decision tree (DT), support vector crnahine (SVМ), k-neearst neigohbur (КNN), nai"ve Bayes (NВ), random forest ()RF and neaurl network (NN). For the experiments, the R statistical envi­ ronment was chosen, as it is an open source scripting language specifically designed ofr data analysis. The classifiers were evaluated based on performance metrics includ­ ing accuracy, sensitivity, and specificity. The SVМ classifier achieved а classification accuracy of 0.9372 with а sensitivity of 0.9332 and а specificity of 0.9226, so it per­ ofrms better than all the remaining classifiers.

А sirnilar method of two-stage data processing is used in [ 44 ] to predict а patient's response to а drug in the treatment of cancer. For prediction of drug response based on molecular profiles of multiple cancer cell types, it was generated а large-scale pharmacogenomics dataset for 1001 cancer cell lines and 251 anti-cancer drugs. The authors performed the feature selection in the form of ASRs and utilized the selected efatures to train the state-of-the-art Deep Leaming Neural Networks (DLNNs) to predict pharmacological response in а Ыind (control) set. The ASRs are treated as а novel meta-dataset. Specifically, the Apriori algorithm was applied to generate а rule­ set, containing all tissue-to-gene, tissue-to-drug, gene-to-drug and drug-to-drug asso­ ciations.

The study shows that type 2 diabetes is а genetic disease and evidence of а statisti­ cal interaction among several Single Nucleotide Polymorphisms (SNPs) has been reported. In [ 26 ], the algorithm Apriori-Gen has been applied to SNP data of type 2 diabetes for association study. The obtained associations are measured through risk rate ()R nad odds ratio (OR) proposed Ьу the authors. The obtained results allow to assess with high accuracy and statistical reliaЬility the interaction of nucleotide poly­ morphisms with disease complexes. An analysis of diabetic ASR based on the rApiori algorithm is given in [ 45 ].

The volume of Ьiological knowledge is riapdly increasing in the rfom of gene ex­ pression databases (GEO, Arrayexpress, etc.), information on rnicroaпay experiments (spotted probes, data processing protocols, etc.), molecular databases (GenВank, ЕmЫ, Unigene, etc.), semantic sources as thesaurs, ontologies or semantic networks (UМLS, GO, etc.), ЬiЫiogrhapical databases (Medline, Biosis, etc.) and gene/protein related specicfi sources (КEGG, OMIM, etc.) [ 27 ]. 4

Perfection of the Apriori Algorithm The popularity of the Apriori algorithm for medical diagnostic tasks is due to its sim­ plicity, however, its application for large data sets requires the development of more eficient modifications in terms of reducing its computational complexity. And such work to improve this algorithm is being carried out all over the world [ 4 ]. 1n particu­ lar, we may Ье аЫе to choose the following directions in improving the Apriori algo­ rit:hm 1) developing new algorithms; 2) data management; 3) constraint-based ASR nrining; 4) incremental mode of ARS rnining.

Developing New Algorithms. The algorithms FP-GROWS and ECLAT are at­ irtbutaЬle to the rfist direction. The main drawbacks of the Apriori algorithm is scan­ ning the database several times. The algorithm FP-GROWTH [ 14, 32 ] uses а :fre­ quent-pattem tree structure (FP-Tree), which stores all the database. This structure can compress the data up to 200 times, and it is stored to the computer's memory. Then, efrquent itemsets are directly extracted from the FP-Tree using the divide-and­ conquer method. This algorithm is used for analyzing risk factors of Туре 2 diabetes in [ 46 ].

The algorithm "Equivalence Class Transformation" (ECLAT) mines frequent item­ sets in а vertical data format [ 12 ]. The algorithm builds the TID set of all items in the rtansaction database. The ASR mining on vertically partitioned data is used in [ 15 ].

The article [ 45 ] gives а general overview of effective processing the rApiori algo­ rithm on medical data. The authors propose а modification of the Apriori algorith щ in which the amount of support for many candidates for frequency multiple attributes is calculated only for transactions, the length of which is longer or equal to the cardi­ nality of the candidates in question. The study in [ 9 ] aims to see the efect of the k­ means clustering algorithm on the Apriori algorithm Ьу comЬining these two algo­ rithms. А logical comЬinatorial neuron-like network is advanced for optimization of the Apriori algorithm in [ 30 ].

Data Management. The following articles can Ье attributed to the category of data management. In [ 37 ], the ASRs are extracted to predict the co-diseases in diabetic mellitus patients. The peculiarity of this work is in selecting rules via their testing on а sample of data not used in data processing. The eficiency of the Apriori algorithm was increased with the help of а prefixed-itemset-based data stcruture [ 49 ]. 1n paper [ 38 ], а new method and а statistic test on rules were introduced to mine ASR over umltiple databases. In [ 42 ], а method for ASRs extraction based on ontology seman­ tics is proposed. The medical dataset is transformed into an ontology in the form of rtiples (subject, object, predicate), and SPARQL (Query Results ХМL Format) is used to query the generated ontology.

Constraint-Based Association Rules Mining. Currently, а lot of studies appeared in iwhch the Apriori algorithm is optimized from the point of view of obtaining not all the possiЫe set of ASRs but only some of its subset satisfying а given essential prop­ erty - interesting rules, non-redundant, negative, maximal rules, association rules generated through the questions of the end users and some others.

The authors of [ 1 ] introduce а Query-constraint based ASR Mining (QARМ) pa­ proach for exploratory analysis of multiple, diverse clinical data sets in the National Sleep Resecarh Resource (NSR). Top-k Non-Redundant (ТNR) ASR mining algo­ rithm is used in this work. Non-redundant ASR is rule deleting from which at least one item implies the change of rule's support for the worse. The work [ 36 ] also pro­ poses an algorithm to generate non-redundant ASRs.

Both positive and negative rules were generated in [ 27 ] to analyze which diagnosis types require or not require Laboratory Diagnostic Tests (LDTs) for patients. The negative rules are generated from infrequent itemsets.

Maximal ASRs are extracted rfom maximal frequent pattems (itemsets). А pattem Х is а xmaimal rfequent in data set D if Х is frequent, and there exists no super­ pattem х such that Х с х and х is efrquent in D. The maximal frequent pattem mining algorithms is, for example, Maximal Frequent Itemset Algorithm (MAFIA) [ 14, 21 ].

Currently there is consideraЫe interest in the methods restricting the extraction of rules to the specicfi type of the most interesting rules for the users. In [ 33 ], it is advanced а method ofRank Based Weighted ASR Mining (RANWAR)) elaborated for Ьiological data processing. Two new measures of ientrestinegnss era considered based on ranging gens.

In [ 43 ], а method for generating efficient rules for associative classification is ad­ vcaned. Associative classicfiation is а technique that integrates classicfiation and ASRs mining for classifying unseen data. Associative classification gives more accu­ rate dan easier to dunerstand rules than it is possiЫe to obtain Ьу using the traditional classiefirs. In [ 37, 40 ], mining ASRs is also combined with extracting classification rules.

Existing ASRs mining algorithms rely on frequency-based rule evaluation methods afiling to provide sound statistical or computational measeurs for rule evaluation, and oeftn sufer from many redundant rules. In [ 37 ], the authors propose predictaЬility­ based an ASRs mining algorithm based on cross-validation with а new rule evaluation step. А training dataset is partitioned into inner training sets and inner test sets and then candidate rules' predictive performance is evaluated.

lncremental Mode of Association Rules Mining Traditional static ASRs mining cannot solve real-world proЫems with dynamically changing data. eWhn the size of the transaction database increases, then an initially frequent item may become an inefrquent one, and an initially infrequent item may become а frequent one. Incre­ mental ASRs mining algorithms help to сору with the drawbacks of the classical Apriori algorithm. The authors [ 52 ] comЬine the Fast Update Priunng (FUP) algo­ rithm with а compressed Boolean matrix [ 25 ] and propose а new incremental ASRs mining algorithm, named FBCM. This algorithm requires only а single scan of both the database and incremental database. An incremental algorithm for mining interest­ ing ASRs has been developed in [ 48 ]. The papers [ 50 ] summarize the methods for incremental ASRs mining.

Conclusion The paper provides an overview on mining associative rules from data in Ьiomedical research. This review is based on а study of the work from 2013 to 2020 and shows the widespread use of the Apriori algorithm and its modifications in medicine. The review includes also the methods to improve the Apriori algorithm to mining more eefctive associative rules adapted for various research tasks.

References

1. Abeysinghe , R. and Cui , L. : Query-constraint-based mining of association rules for exploratory analysis of clinical datasets in the National Sleep Research Resource . Medical Inanfotics and Decision Making 18 ( Suppl 2 ), 89 - 100 ( 2018 )

2. Aagrwal , R. , Imielinski, Т., and Swami, А.: Mining association rules between sets of items in large databases . АСМ SIGMOD Record , 22 ( 2 ), 207 - 216 ( 1993 )

3. Altaf , W. , Shahbaz, М., dan Guergachi , А.: Applications of association rule mining in health infotnaics: А survey . tAricfiial Intelligence Review , 47 ( 3 ), 313 - 340 ( 2017 )

4. Bhende , D. , Kaskarer , U. , and Gedaщ М.: Study of various improved Apriori algorithm . IOSR Journal ofComputer Science Engineering , e-ISSN: 2278 - 0661 , рр. 55 - 58 ( 2016 )

5. Billig , V. А ., and Ivanova, О.У.: Building association rules in а task of medical diagnosis . Software Products and Systems , 2 ( 114 ), 146 - 157 ( 2016 ) (in Russian) .

6. Birukov , А. Р. dan uDrnasky, S. М.: Revealing the comorbidity of professional diseases caused Ьу pathogenic factors with eth help of associative algorithms on eth expamles of а cohort of victims of ionizing radiation . Medicine of Extreme Situations , 2 , 13 - 24 ( 2016 ) (in Russi)an .

7. Chavez , R. , Gбrriz , J. М. , Rameirz , J. , Salas-Gonzlez , D. , and Gбmez-Rio , М.: Ecfiient mining of association rules rfo the early diaognsis of Alzheimer's disease . PhisMed Biol ., 56 , 6047 - 6063 ( 2011 )

8. Das , N. N. et al.: Brief survey on DNA sequence mining . Intemational Journal ofComputer Science dan Mobile Computing , 2 ( 11 ), 129 - 134 ( 2013 )

9. Dharshini , N. Р. , Azmi , F. , Fawwaz , I. , Husein, А. М., and Sireg,ar S. D.: Analysis of accuracy k-means dan Apriori algoritshm for patient data clusters . Journal Physics.:Conference Ser . 1230 , 1 - 9 . ЮР PuЫishing ( 2019 )

10. Dhumale S.: Predicting pattems over protein sequences using Apriori algorithm . Intemational Joaurnl of Engineering andComputer Science 4 ( 7 ), 13011 - 13016 ( 2015 )

11. Doddi , D. , Marth, А., Ravi , S.S. , and Tomey, DC.. : Discovery of association rules in medical data . Medical Inarmfotics dna Intemet inMedicine 26 ( 1 ), 25 - 33 ( 2001 )

12. Domadia , N. and Rao , U.P. : Privacy-preserving disitrbuted association rule mining approach on vertically partitioned healthcare data . ProcediaComputer Science , 148 , 303 - 312 ( 2019 )

13. Go , Е., Lee , S. , and Yoon, Т.: Analysis of ebolavirus with decision etre and Apriori algorithm . Intemational Journal ofMachine Leaming andCoumpting , 4 ( 6 ), 543 - 548 ( 2014 ).

14. Han , J ., Pei , J. , and Yin, У.: Mining frequent pattems without candidate generation . In: Proceedings of АСМ SIGMOD Intemational Conference on Management of Data, рр. 1- 12 ( 2000 )

15. Harahap , М., Husein , А.М., Aisyah , S. et al.: Mining association rules based on diseases population for recommendation of medicine need . Journal of Physics: Conference Series , 1007 , 1 - 12 ( 2018 )

16. Jalali , S. М. , Moro , S. , Mahmoudi, М. R. , Ghaafry, К. А., Maleki , М., and Alidoostan, А.: А comparative naalysis of classifiers in cancer prediction using multiple data mining techniques . Intemational Joaurnl of Business Intelligence dan Systems Engineering , 1 ( 2 ), 166 - 178 ( 2017 )

17. Kalaiyarasi , R. and Prabasri , S. : Predicting eth Lung Cancer rfom Biological Sequences . Intemational Joaurnl of Innovation in Engineering and Technology (IJIET) , 5 ( 1 ), 106 - 111 ( 2015 )

18. Kanageswari , S. dan Gladis , D.: Generation of association rules of data mining for lung cancer Ьу air pollution . Intemational Journal of Engineering dan Advanced Technology (IJEAT) , 9 ( 3 ), 2874 - 2880 ( 2020 )

19. Kang 'ethe, S. М . dan Wagacha, Р. W.: Extracting diagnosis pattems in electronic medical records using association rule mining . Intemational Journal of Computer Applications 108 ( 15 ), 19 - 27 ( 2014 )

20. Karaolis , М., Moutris , J.A. , Papaconstantinou , L. , and Pattichis, С. S.: Association rule analysis for the assessment of the risk of coronary heart events . In: Proceedings of the 31st nAual Intemiatonal Conference of the IEEE Engineering inMedicine and Biology Society , рр. 6238 - 6241 .Minneapolis, МN: USA ( 2009 )

21. Кavakiotis , I. , Tzanis , G. , and Vlahavas , I.: Mining efrquent patterns and association rules rfom biological data . In: Mourad Ellourni dna Albert У. Zomaya (eds.) , Biological nКowledge Discovery Handbook: Preprocessing , Mining, and Posrtpocessing of Biological Data, First Edition , Chapter 34 , рр. 737 - 762 . John Wiley & Sons, Inc. PuЫished ( 2014 )

22. Kumar , R. Р. R. , Jay,akumr R., and Sankridevi, А.: Apriori-based rfequent symptomset association mining in medical databases . International Joaurnl of Recent Technology and Engineering (IJRТE) , 7 ( 5С ), 65 - 68 ( 2019 )

23. Lakshmi , К.R. , Кrishna , M.V. , dan Kumar , S.P. : Performance comparison of data mining techniques for predicting of heart disease survivability . Intemational Joaurnl of Scientific and Research PuЬlications , 3 ( 6 ), 1 - 10 ( 2013 )

24. Lee , J.Y. and Кim , К.-У.: Semantic dan association rule mining-based knowledge extension for reusaЫe medical equipment random forest rules . Journal oflntegrated Design and Process Science , 22 ( 4 ), 55 - 81 ( 2018 )

25. Li , Т. dan Luo , D.: А new improved Apriori algorithm based on compression matrix . In: Proceedings of Intemational Conference on Advanced Data Mining Application. , vol. 8933 , рр . 1- 15 . Berlin, Germany: Springer-Verlag ( 2014 )

26. Мао , W. and Мао , J.: Тhе Application of Apriori-Gen algorithm in the association study in type 2 diabetes . In: Proceedings of 3rd International Conference on Bioinfoarmtics and Biomedical Engineering (ICBBE) , 4 ( 1 ), 126 - 140 ( 2009 )

27. Martinez , R. , Pasquier, С., and Pasquier , N.: GENМIER: Mining informative association rules from genomic data . In: Proceeding of the IEEEIntemational Conference on Bioinformatics dan Biomedicine , рр. 15 - 22 ( 2007 )

28. Mayilvagaann , М. dan Hemalathe, R.: Performance comparison of FSA Red and Apriori algorithm's in mutation analysis . International Joaurnl of Computer Trends and Technology (IJСТТ) , 17 ( 4 ), 205 - 209 ( 2014 )

29. Mirmozafari , М., Alinezhad , А., and Gilanpour, А.: Data mining Apriori algorithm rfo heart disease prediction . tInemaotinal Joaurnl of Computing , Communication & Instrumentation Engineering (IJCCTE), 4 ( 1 ), 20 - 23 ( 2017 )

30. Naidenova , Х., Parkhomenko , V. , and Shvetsov, К.: Application of а logicalcombinatorial network to symbolic machine learning tasks . In: Naidenova, Х., Shvetsov , К., and Yakovlev, А. (eds.), Machine Learning in Analysis of Biomedical and SocioEconomicData , рр. 371 - 409 . Saint-Petersburg, Russia: Polytech Press ( 2020 )

31. Nahar , J. , Tickle, К., Ali , A.B.M.S. , and Chen, У.Р. Significant ccaner prevention factor extraction: An association rule discovery approach . Joaurnl of Medical System 35 ( 3 ), 353 - 367 ( 2011 )

32. Pinheiro F.M.R.: Applying the Apriori dan FP-Growth association algorithms to liver cancer data . А Thesis Submitted in Partial Fulfillment of the Requirement for the Degree of Master of Science , 172 рр. University of Victoria ( 2013 )

33. Premalatha S. and Nandhini , U. С.: Efficiently generating the rank based weighted association rule mining using pAriori algorithm in High Biological Database . Intemational Resecarh Journal ofEngineering dan Technology (IRJEТ) , 2 ( 9 ), 2143 - 2147 ( 2015 )

34. Ramezankhani , А., Pournik , О., Shahrabi , J. , Azizi , F. , and Hadegh , F. : An application of association rule mining to extract risk pattem for type 2 diabetes using tehran lipid and glucose study database . Intemational ofEndocrinol Metabolism . 13 ( 2 ) е25389 ( 2015 )

35. Sanyer , G. and Ta ar, С. 6.: Highlighting the rules between diagnosis types and laboratory diagnostic tests for patients of an emergency department: use of association rule mining . Health Informatics Journal , 2 ( 6 ), 1 - 17 ( 2019 ).

36. Severac , F. , Sauleau , Е. А., Meyer, N., Lefevre , Н., Nisand , G. , and Jay , N.: Nonredundtan association rules between diseases and medications: na automated method rfo nkowledge base construction . ВМС MedicalInfoпnatics dan Decision Making 15 ( 1 ), 2 - 7 ( 2015 )

37. Shahebaz , Ah. Кh. and Jabbar , М. А.: Improved classification techniques to predict eth codisease in diabetic mellituspatients using discretization dan Apriori algorithm . IntemationalJoaurnl of Innovative Technologyand ExploringEngineering (IЛТЕЕ) , 8 ( 11 ), 730 - 733 ( 2019 )

38. Shang , Е., Du , an J. , Fan, Х., Tang , У., and Уе , L.: Association rulemining and statistic test over multipledatasets on ТСМ drug pairs . IntemationalJournal ofBiomedicalData Mining 6 ( 1 ), 2 - 6 ( 201 7)

39. Shaaпn , N. and Om , Н.: Earlydetection and prevention of oralcancer: association lrues mining on investigation . WSEAS Transaction on Computers , 13 , 1 - 8 ( 2014 )

40. Song , К. dan Lee, К.: Predictability-basedcollectivcelassassociation rulemining . Expert System with Application , 79 , 1 - 7 ( 2017 )

41. Swathi , Р. dan Prajna, В.: Тhе efective procession of Apriori algorithmprescribed data mining on medical data . Intemational Joaurnl of Computer Science and Technology (IJCST) , 7 ( 3 ), 22 - 26 ( 2016 )

42. Тhamer , V. , El-Sappagh , S. , dan El-Shishtawy, Т.: А semantic approach for extracting medicalassociation rules . Intemational Joaurnl oflntelligenEtngineering Systems , 13 ( 3 ), 280 - 293 ( 2020 )

43. Тhaanjiranthom , Ch. and Songram , Р.: Generation of efficient rulesrfo associative classiifcation . In: Proceedings of the 13 th IntemationalConference MIWAl , рр. 109 - 120 ( 2019 )

44. Vougas , К., Кrochmal , М. , Jackson , Т. , Po lyz os, А . et al: Deep leamingand association rulemining rfo predicting gdru response in cancer. А personalized medicine paproach . Cuпent version availaЫehttps://ww.Ьiorxiv.org/content/10.l l01/070490v3.lful ( 2017 )

45. Wang , Х., Su , К., and Liu, Zh.: Analysis of diabetic association rulesbased on Apriori algorithm . In: Ch. Huang, Yu-Wei Chan , nad NeilYen (eds), Data Processing Techniques and Applicationsfor Cyber-PhysicalSystem, рр. 553-563 . Springer ( 2020 ).

46. Wei , Z. , and Guigjan ,an Уе.: Тhе research on analyzingrisk factors of type 2 diabetes mellitubsased on improved tfequent pattem tree algorithm . 1n Proceedings of the 2015 IntemalConference on MaterialsEngineering dan Infoпnation TechnologyApplications , рр. 459 - 463 . Atltanic Press ( 2015 )

47. Win , S.L. , Htike , Z.Z. , Yusof , F. , and Noorbatcha , 1.А.: Gene expression mining for survivaЬilityof patients in earlystages of lungcancer . lntemational Journal ofBioinпnfoatics andBiosciences , 4 ( 2 ), 1 - 9 ( 2014 ).

48. Ya ,fi Е., Al-Hegami , А., Alam ,А., dan Biswas , R. УАМI: IncrementalMining oflnteresting Association pattems . Тhе IntemationalArab Joumaloflnfoпnation Technology9(6) , 504 - 5011 ( 2012 )

49. Yu , S. dan Zhou , У.: А prefixed-itemset-based improvement rof Apriori algorithm . CornellUniversity , arXiv preprint, arXiv: 1601.01746 ( 2016 )

50. Zhan , F. , Zhu , Х., Zhang , L. , Wang , Х., Wgan , L. , and Liu, Ch.: Surnmay of association rules . ЮРConf. Series: Earth dna Environment Science (ЕSМА) , vol. 252 , рр. 12 ( 2019 ).

51. Zhang , B.Z. , Jiang , К. Q. , and Zhang, У. Z.: Survey on incrementalassociation rulemining research . Journal of China Computer. Systems , 37 ( 1 ), 18 - 23 ( 2016 )

52. Zhou , D. , Ouyag , V. , Kgan , Zh. , Li , Zh. , Zhou , J.P. , dan Cheng, Х.: lncrementalМАR based on maitrx cormpession ofr edge computing . IEEE Access , Section on Innovation and Application of IntelligenPtrocessing of Data, Infoпnation nad nКowledge as Resources in Edge Computing , vol. 7 , 173044 - 173053 ( 2019 )