IoT-Cloud-Based Smart Healthcare Monitoring System for Heart Disease Prediction via Deep Learning

Ci a ion: Nancy, A.A.; Ra ind an, D.; Raj Vincen , P.M.D.; S ini asan, K.; Gu ie ez Reina, D. IoT-Cloud-Based Sma Heal hca e Moni o ing Sys em o Hea Disease P edic ion ia Deep Lea ning. Elec onics 2022,11, 2292. h ps://doi.o g/10.3390/ elec onics11152292 Academic Edi o s: F ancisco Luna-Pe ejón, Lou des Mi ó Ama an e and F ancisco Gómez-Rod íguez Recei ed: 20 June 2022 Accep ed: 19 July 2022 Published: 22 July 2022 Publishe ’s No e: MDPI s ays neu al wi h ega d o ju isdic ional claims in published maps and ins i u ional a il- ia ions. Copy igh : © 2022 by he au ho s. Licensee MDPI, Basel, Swi ze land. This a icle is an open access a icle dis ibu ed unde he e ms and condi ions o he C ea i e Commons A ibu ion (CC BY) license (h ps:// c ea i ecommons.o g/licenses/by/ 4.0/). elec onics A icle IoT-Cloud-Based Sma Heal hca e Moni o ing Sys em o Hea Disease P edic ion ia Deep Lea ning A Angel Nancy 1, Dakshanamoo hy Ra ind an 1, P M Du ai Raj Vincen 2, Ka hi a an S ini asan 3 and Daniel Gu ie ez Reina 4,* 1Depa men o Compu e Science, S . Joseph’s College (Au onomous), Bha a hidasan Uni e si y, Ti uchi appalli 620002, India; [email p o ec ed] (A.A.N.); [email p o ec ed] (D.R.) 2School o In o ma ion Technology and Enginee ing, Vello e Ins i u e o Technology, Vello e 632014, India; [email p o ec ed] 3School o Compu e Science and Enginee ing, Vello e Ins i u e o Technology, Vello e 632014, India; [email p o ec ed] 4Elec onic Enginee ing Depa men , Uni e si y o Se ille, 41004 Se illa, Spain *Co espondence: dgu ie [email p o ec ed] Abs ac : The In e ne o Things con e s seamless connec i i y be ween people and objec s, and i s con luence wi h he Cloud imp o es ou li es. P edic i e analy ics in he medical domain can help u n a eac i e heal hca e s a egy in o a p oac i e one, wi h ad anced a i icial in elligence and machine lea ning app oaches pe mea ing he heal hca e indus y. As he sub ield o ML, deep lea ning possesses he ans o ma i e po en ial o accu a ely analysing as da a a excep ional speeds, elici ing in elligen insigh s, and e icien ly sol ing in ica e issues. The accu a e and imely p edic ion o diseases is c ucial in ensu ing p e en i e ca e alongside ea ly in e en ion o people a isk. Wi h he widesp ead adop ion o elec onic clinical eco ds, c ea ing p edic ion models wi h enhanced accu acy is key o ha nessing ecu en neu al ne wo k a ian s o deep lea ning possessing he abili y o manage sequen ial ime-se ies da a. The p oposed sys em acqui es da a om IoT de ices, and he elec onic clinical da a s o ed on he cloud pe aining o pa ien his o y a e subjec ed o p edic i e analy ics. The sma heal hca e sys em o moni o ing and accu a ely p edic ing hea disease isk buil a ound Bi-LSTM (bidi ec ional long sho - e m memo y) showcases an accu acy o 98.86%, a p ecision o 98.9%, a sensi i i y o 98.8%, a speci ici y o 98.89%, and an F-measu e o 98.86%, which a e much be e han he exis ing sma hea disease p edic ion sys ems. Keywo ds: cloud compu ing; In e ne o Things; heal hca e; p edic i e analy ics; ecu en neu- al ne wo k 1. In oduc ion Human e olu ion has un olded in syne gy wi h science and echnology e olu ion. In o ma ion and communica ion echnology (ICT) ad ancemen s ha e laid he ounda ion o inno a i e solu ions in di e se indus y domains, including heal hca e, ag icul u e, anspo a ion, and logis ics, among o he s. The In e ne o Things (IoT) is a subs an i e d i ing o ce o ICT echnological ad ancemen , leading p ospec i e sec o s down he oad o au oma ion alongside decen alized in elligence [ 1 ]. The IoT e ol es incessan ly and impac s e e y ace o ou li e while esembling a li ing en i y. F om household appliances o obo s in ac o ies, he IoT connec s da a, people, hings/objec s, and p ocesses. Mean- while, cloud compu ing (CC) deli e s on-demand elas ic se ices wi h i ually unlimi ed compu a ion and s o age capabili y [ 2 ]. Despi e being unique and independen in hei espec i e e olu ion, cloud compu ing and IoT aspec s complemen one ano he . E en u- ally, he wo echnologies con e ged in ecen yea s, and he con luence became known as a Cloud–IoT pa adigm [ 3 , 4 ], o e ing emendous p ospec s o d i ing new inno a i e se ices and applica ions. Elec onics 2022,11, 2292. h ps://doi.o g/10.3390/elec onics11152292 h ps://www.mdpi.com/jou nal/elec onics Elec onics 2022,11, 2292 2 o 19 Heal hca e- ela ed applica ions ha e pushing o inno a ions in science and echnol- ogy since in o ma ion echnology applica ions began o emo ely acqui e, ack and con ol he s a us o pa ien s. Thus, IoT d i es ecen inno a ions in heal hca e and e olu ionizes hem by acqui ing he physiological da a o pa ien s h ough senso ne wo ks and wea able de ices [ 5 ]. The Cloud–IoT ha nesses he Cloud’s emendous po en ial o he s o age and p ocessing o an eno mous olume o clinical eco ds o pa ien s including senso da a om medical IoT o heal hca e analy ics. Analy ics ensues he sys ema ic quan i a i e and quali a i e analysis o conce ned da a o e icien decision making, while p edic i e analy ics s ems om ad anced analy ics aiming o elici he p ognosis o u u e occu ences using he a ailable da a [ 6 ]. The analy ics in heal hca e is ha nessed o clinical decision suppo , p edic i e isk assessmen , and emo e heal h moni o ing, among o he c ucial asks. P edic ing and lowe ing isk based on cu en and pas pa ien da a a e a big pa o medicine. The in eg a ion o humongous da a om dispa a e sou ces comp ising elec onic heal h eco ds, medical imaging, sc eening esul s, and adminis a i e in o ma ion wa an ing swi decisions is e ec i ely ackled by heal hca e analy ics [ 7 ]. Clinicians mus o en make decisions wi h a high deg ee o unce ain y; howe e , wi h he headway o p edic i e analy ics in heal hca e, hose decisions will be mo e in o med han e e . These cu ing-edge p edic i e analy ics app oaches help iden i y ouble ea ly on, a oid complica ion isks, imp o e ch onic illness managemen , e ade hospi al eadmission, ecei e medical esea ch aid, and minimize o e head expenses. P edic i e analy ics in heal hca e deploys di e se echniques om con en ional lin- ea models o ad anced algo i hms o a i icial in elligence (AI) and machine lea ning (ML) [ 8 ]. Deep lea ning (DL), a sub ield o ML, is su icien ly eliable and obus o au- oma ically handle and lea n om a huge amoun o complex heal hca e da a and o e s ac ionable insigh s and solu ions o in ica e p oblems. I s deploymen o a wide a ie y o medical applica ions has su passed he esul s o adi ional models. Speci ically, he ecu en neu al ne wo k (RNN) [ 9 ] is compe en a managing he long- e m dependen- cies o inpu da a and has g own p ominen in he s udy o empo al e en s conce ning ime-sequen ial applica ions. 1.1. Mo i a ion P edic i e analy ics is p o ing i s wo h, no jus in he hospi al en i onmen , bu also a home by emo e moni o ing and keeping pa ien s om elapsing in o he need o acu e ea men . P edic i e analy ics aid in he diagnosis, p ognosis, and he apy a e e y s age o a pa ien ’s ea men [ 10 ]. I also helps in designing he ea men cou se, p o iding clinical decision suppo , dec easing ad e se occu ences, and enhancing he o e all ca e quali y while lowe ing heal hca e cos s. Mo eo e , he pe sonalized heal hca e model shi s om ea ing pa ien s as numbe s o ea ing hem as indi iduals, cus omizing ea men o hei unique medical his o y, en i onmen , social isk ac o s, gene ics, and biochemis y, among o he hings [ 11 ]— a he han depending on demog aphic s a is ics ha do no apply o e e yone. I ende s eal- ime clinical decision assis ance a he poin o ea men , allowing o he mos e icien deli e y o indi idualized heal hca e [ 12 ]. Wi h deadly diseases, spo ing hem ea ly on and de ec ing any possible de e io a ion in he pa ien s’ condi ion be o e occu ence can signi ican ly imp o e he odds o an e ec i e ea men . The diseases a ec ing he hea and i s ela ed blood essels a e all classi ied as ca dio- ascula diseases (CVDs). These include a hy hmia, co ona y a e y disease, congeni al hea disease, al e disease, ao ic disease, hea ailu e, pe iphe al a e y disease, pe ica - dial disease, hea al e disease, ce eb o ascula disease, heuma ic hea disease, deep ein h ombosis, ca diomyopa hy, myoca di is, a ial ib illa ion, ischemic hea disease, and s oke [13–15]. The mos p ominen cause o global mo ali y is ca dio ascula diseases (CVDs), claiming he li es o an es ima ed 17.9 million indi iduals and accoun ing o 32% o all a ali ies wo ldwide [ 16 ]. Hea a acks and s okes cause ou ou o e e y i e CVD Elec onics 2022,11, 2292 3 o 19 a ali ies, which a e 85% o all CVD mo ali ies, wi h one- hi d occu ing be o e age 70. Iden i ying indi iduals a isk o CVDs and ensu ing ha hey ecei e p ope he apy can help a e un imely dea hs. This is whe e he p edic i e algo i hms powe ed by AI and ML come in o play alongside he In e ne o Things, as hese a e adep a managing massi e and di e se da a. Pa e n classi ica ion, as a pa e n ecogni ion ask, is a c ucial supe ised lea ning pa adigm o iden i ying and classi ying disease pa e ns in he medical ield [ 17 ]. The esea che s wo king on classi ica ion algo i hms conce ning hea disease s i e o achie e he maximum classi ica ion accu acy possible as pa ien s’ li es a e a s ake. Many indi iduals a e a isk o hea disease due o long- e m condi ions such as pe sis ing high blood p essu e. Wi h he inc ease in he aging popula ion ac oss he globe, mos o hem a e diagnosed wi h ch onic hea condi ions. This wa an s he con inuous eal- ime moni o ing o indi iduals a in-home ca e and he pa ien s in ea men wi hin hospi al p emises, en ailing imely ea men upon he luc ua ion o i al signs. The p olonged acking o heal h condi ions in he elde ly helps minimize hospi aliza ion cos and enhance he quali y o li e, bu con en ional me hods a e edious and i ing. This necessi a es e icien acili ies o mi iga e he o e whelming wo kload o clinicians and hospi al s a while minimizing he cos o heal h moni o ing. The pe asi e na u e o IoT has inci ed he p oli e a ion o sma , in e connec ed de ices and wea ables wi h senso s, he eby acili a ing emo e pa ien moni o ing pe aining o hea disease. The IoT o heal hca e moni o ing includes sma heal h wa ches, wea able blood p essu e moni o s, and wea able ECG moni o s equipped wi h medical senso s. Thus, he heal hca e IoT acqui es i al pa ien da a and ansmi s hem o he Cloud o s o age and complex deep lea ning analy ics along wi h p io elec onic clinical eco ds o accu a e hea isk diagnos ics. These IoT de ices can swi ly no i y he clinicians and ca e ake s o he pa ien ’s condi ion. This enables clinicians o be e make imely decisions o indi iduals as well as he popula ion a la ge by es ima ing pa ien s’ chance o de eloping a speci ic hea disease, hei p ognosis o he gi en condi ion, and he co esponding ea men . 1.2. Con ibu ion The pi o al ou comes o his esea ch ini ia i e a e lis ed as ollows: 1. The da a collec ed om IoT senso s pe aining o hea disease isk p edic ion a e subjec o he da a p e-p ocessing asks o da a cleaning and da a il e ing a he Cloud laye ; 2. The ensuing da a a e sen o he uzzy in o ma ion sys em (FIS) o he ini ial classi i- ca ion ask; 3. Finally, he p oposed Bi-LSTM model is used o accu a ely p edic he isk o hea disease in pa ien s. The emaining sec ions o his a icle a e o ganized in o ela ed wo k, me hodol- ogy, expe imen al se up, pe o mance assessmen , expe imen al esul s and discussion, compa a i e analysis, u u e di ec ions, and conclusions. 1.3. Rela ed Wo k In ecen imes, di e se sys ems o hea disease p edic ion ha e been p opounded. Fo enhancing hea disease isk p edic ion accu acy, he deploymen o se e al ensemble classi ie s displays an accu acy o 85.4% [ 18 ]. A model o diagnosing hea disease diagnosis ha combines ough se s-based a ibu e educ ion in ol ing he chaos i e ly algo i hm wi h an in e al ype-2 uzzy logic sys em showcases an accu acy o 86% [ 19 ]. A machine lea ning hyb id model o p edic hea disease [ 20 ] by combining andom o es (RM) wi h linea me hod (LM) app oaches exhibi s a pe o mance accu acy o 88.7%. An in eg a ed decision suppo sys em o p edic ing he isk o hea ailu e, which combines a uzzy analy ic hie a chy p ocess and a i icial neu al ne wo k o ea u e weigh - ing and classi ica ion asks, espec i ely, achie es 91.0% accu acy [ 21 ]. A sma sys em o diagnosing hea disease deploying a χ2 s a is ical model and a deep neu al ne wo k o Elec onics 2022,11, 2292 4 o 19 ea u e e inemen and classi ica ion asks, espec i ely, is p oposed. The model a ains an accu acy, speci ici y, and sensi i i y o 91.57%, 93.12%, and 89.78%, espec i ely [22]. An adap i e weigh ed uzzy ule-based sys em o assessing he hea disease isk le el is p esen ed. This au oma ic diagnos ic sys em o a uzzy model based on a gene ic algo i hm along wi h a deployed modi ied dynamic mul i-swa m pa icle op imiza ion app oach shows an accu acy o 92.3% [ 23 ]. A hea disease iden i ica ion model deploying algo i hms o uni a ia e and elie ea u e selec ion alongside a decision ee o classi ica- ion achie es an accu acy o 92.8% [24]. A co ona y a e y disease p edic ion model wi h a neu o- uzzy medical decision suppo sys em is p esen ed. This sys em in ol es an a i icial neu al ne wo k and an adap i e neu o- uzzy in e ence sys em, which displays an accu acy, sensi i i y, speci ici y, and p ecision o 94.15%, 91.44%, 95.59%, and 92.61%, espec i ely [ 25 ]. A sys em o au oma ically p edic ing hea disease is p oposed, which deploys clus e -based Bi-LSTM (bidi ec ional long sho - e m memo y). When es ed wi h he UCI da ase , his model exhibi s an accu acy o 94.78% [26]. An expe sys em o diagnosing hea disease by combining uzzy ules and deep neu al ne wo ks is p esen ed, showing an o e all accu acy o 96.5% [ 27 ]. A me hod ha in eg a es CNN wi h deep lea ning algo i hms e e ed o as Ca dioHelp is in oduced, which uses CNN o ea ly hea ailu e p edic ion in ol ing a empo al model. This app oach ou pe o ms o he s a e-o - he-a me hods, wi h a 97% accu acy a e [28]. An IoT-based hyb id sys em o ca dio ascula disease p edic ion is o e ed, including sequen ial o wa d selec ion (SFS) as he ea u e selec ion echnique and a andom o es o classi ica ion. This sys em ecommends physical as well as die a y plans o pa ien s in unc ion o hei age and gende , showing 98% accu acy compa ed o o he heu is ic model ecommende sys ems [ 29 ]. A model capable o handling medical da a om mul iple senso s in ol ing an ensemble classi ie —Ke nel andom o es [ 30 ]—shows 98% accu acy when deployed on a hea disease da ase . A new IoT amewo k based on deep con olu ional neu al ne wo ks, which a e con- nec ed o a wea able senso ha measu es he blood p essu e and ECG o a pa ien , is sugges ed. When compa ed o logis ic eg ession and exis ing deep lea ning neu al ne - wo ks, his echnique pe o ms be e wi h 98.2% accu acy [ 31 ]. A sma sys em p edic ing he isk o hea disease [ 32 ] om da a acqui ed by wea able senso s and pa ien medical his o y, based on he ensemble deep lea ning model Logi boos along wi h ea u e usion echnique, was p esen ed. The sys em shows 98.5% accu acy in hea disease diagnoses while au oma ically ecommending die a y plans in unc ion o he heal h condi ion. A model o p edic ing hea disease is p esen ed by combining he me hod o embed- ded ea u e selec ion in ol ing he Linea SVC algo i hm wi h deep neu al ne wo ks. This sys em achie es an accu acy, ecall, p ecision, and F-measu e o 98.56%, 99.35%, 97.84%, and 98.3%, espec i ely, when e alua ed wi h he hea disease da ase [33]. The a o emen ioned s a e-o - he-a wo ks pe aining o hea disease isk diagnosis ha nessing he UCI hea disease da ase p edominan ly a ail s a is ical and machine lea ning algo i hms and me hods. The classi ica ion accu acy shown by he exis ing me hods has he possibili y o u he enhancemen when deep lea ning app oaches a e emphasized. Mo eo e , u ilizing he uzzy sys ems alongside ecu en neu al ne wo k algo i hms has he po en ial o o e be e ou comes. The ensuing sec ions elabo a e on he p oposed uzzy-based ecu en neu al ne wo k model o accu a e hea disease isk p edic ion. 2. Ma e ials and Me hods IoT echnology ac s as he c i ical acquisi ion componen o innume able eal- ime applica ions ha p omo e objec –indi idual in e ac ion. The massi e amoun o da a gene a ed by IoT de ices poses a signi ican challenge o he heal hca e sys em pe aining o he p ocessing, s o age, and managemen o da a. Elec onics 2022,11, 2292 5 o 19 The p oposed sma heal hca e sys em o hea disease isk p edic ion includes modules such as (1) he da a acquisi ion/collec ion laye ; (2) da a p e-p ocessing; and (3) he disease p edic ion laye , which is depic ed in Figu e 1. Elec onics 2022, 11, x FOR PEER REVIEW 5 o 20 p oposed uzzy-based ecu en neu al ne wo k model o accu a e hea disease isk p e- dic ion. 2. Ma e ials and Me hods IoT echnology ac s as he c i ical acquisi ion componen o innume able eal- ime applica ions ha p omo e objec –indi idual in e ac ion. The massi e amoun o da a gene a ed by IoT de ices poses a signi ican challenge o he heal hca e sys em pe ain- ing o he p ocessing, s o age, and managemen o da a. The p oposed sma heal hca e sys em o hea disease isk p edic ion includes modules such as (1) he da a acquisi ion/collec ion laye ; (2) da a p e-p ocessing; and (3) he disease p edic ion laye , which is depic ed in Figu e 1. Figu e 1. Hea disease isk p edic ion sys em—block diag am. 2.1. Da a Acquisi ion/Collec ion Laye The p opounded heal hca e sys em acqui es da a om wo p ima y da a sou ces. The physiological da a o pa ien s such as hei blood p essu e (BP), hea a e, blood suga /glucose le el, espi a ion a e, blood oxygen, choles e ol le el, ac i i y, elec oca - diog am (ECG), elec omyog am (EMG), and elec oencephalog am (EEG) a e ga he ed om he pa ien ’s ou ine heal h moni o ing. These da a a e ansmi ed h ough Blue- oo h/Zigbee o ela ed emo e ga eway de ices and hen o he cloud da a cen e , whe e da a p e-p ocessing and disease p edic ion akes place. The o he da a sou ce is he elec- onic clinical da a (ECD), which comp ise he pa ien ’s medical his o y (including hei his o y o smoking and diabe es), obse a ion epo s, and comp ehensi e clinical (lab) epo s which o e aluable in o ma ion on disease p edic ion and a e s o ed in a cloud da abase. 2.1.1. Da ase Fo he expe imen , o de ec he p esence o hea disease om hea pa ien da a, he Cle eland and Hunga ian da ase om he UCI machine lea ning eposi o y a e conside ed. The p oposed algo i hm was deployed on a hea da ase ha includes 14 a ibu es, as depic ed in Table 1. Table 1. A ibu e desc ip ion. Figu e 1. Hea disease isk p edic ion sys em—block diag am. 2.1. Da a Acquisi ion/Collec ion Laye The p opounded heal hca e sys em acqui es da a om wo p ima y da a sou ces. The physiological da a o pa ien s such as hei blood p essu e (BP), hea a e, blood suga /glucose le el, espi a ion a e, blood oxygen, choles e ol le el, ac i i y, elec oca - diog am (ECG), elec omyog am (EMG), and elec oencephalog am (EEG) a e ga he ed om he pa ien ’s ou ine heal h moni o ing. These da a a e ansmi ed h ough Blue- oo h/Zigbee o ela ed emo e ga eway de ices and hen o he cloud da a cen e , whe e da a p e-p ocessing and disease p edic ion akes place. The o he da a sou ce is he elec onic clinical da a (ECD), which comp ise he pa ien ’s medical his o y (including hei his o y o smoking and diabe es), obse a ion epo s, and comp ehensi e clinical (lab) epo s which o e aluable in o ma ion on disease p edic ion and a e s o ed in a cloud da abase. Da ase Fo he expe imen , o de ec he p esence o hea disease om hea pa ien da a, he Cle eland and Hunga ian da ase om he UCI machine lea ning eposi o y a e conside ed. The p oposed algo i hm was deployed on a hea da ase ha includes 14 a ibu es, as depic ed in Table 1. Elec onics 2022,11, 2292 6 o 19 Table 1. A ibu e desc ip ion. S. No. A ibu e Desc ip ion 1 age Age o pa ien in yea s 2 sex 1 = male; 0 = emale 3 cp Type o ches pain (1 = angina, 2 = a ypical o m o angina, 3 = non-angina, 4 = no symp oms o angina) 4 es bps Res ing blood p essu e 5 chol Choles e ol alue 6 bs Fas ing blood suga alue >120 mg/dL (1 = ue and 0 = alse) 7 es ecg Value o ECG a es (0 = no mal, 1 = abno mal (ST-T wa e), 2 = de ini e en icula ) 8 halach Maximum hea a e eco ded 9 exang Exe cise induced angina (1 = yes; 0 = no) 10 oldpeak Exe cise induced ST Dep ession 11 slope Slope o T segmen peak exe cise (1 = unsloping, 2 = la , and 3 = down sloping) 12 ca Majo essels numbe (0–3) colou ed by luo oscopy 13 hal 3 = no mal; 6 = ixed de ec ; 7 = e e sable de ec 14 a ge The p edic ed hea disease s a us (0 = No and 1 = Yes) 2.2. Da a P e-P ocessing Laye Da a p e-p ocessing has become a equisi e o ML algo i hm deploymen as eal- wo ld da a a e p one o being inconsis en , incomple e, and noisy. E icien hea disease p edic ion om he hea disease da ase equi es missing da a handling, no maliza ion, and ea u e selec ion. Da a acqui ed om wea able senso s a e impac ed due o signal abe a ions, such as missing alues and noise, causing ha oc in he case o hea disease p edic ion, comp omising he p edic ion accu acy, o yielding an e oneous esul . We u ilize a well-known echnique o il e he da a known as Kalman il e ing [32,34], which e ec i ely elimina es duplica e eco ds, noise, and disc epancies om he da a. Owing o i s simple o m, i equi es low compu a ional powe [ 35 ]. This unsupe ised il e ing algo i hm is specialized o handle as eal- ime senso da a and u nish alues close o ha o he ac ual alues om he senso wi hou noise [ 36 ]. In addi ion o his, we use wo o he unsupe ised il e s in he da a il e ing s age: emo ing useless and eplace missing alues [ 32 ]. Wi h ano he 90% o maximum a iance, he i s il e elimina es i ele an a ibu es. The second il e subs i u es he mean as well as median alues o he exis ing da a o any alues missing in he s uc u ed da ase . Fuzzy In e ence Sys em The e m uzzy e e s o some hing as inexplici o ague, and he uzzy sys em is inspi ed by he equisi e o model inhe en ly ague eal-wo ld e en s [ 37 ]. The s anda d uzzy sys em is cha ac e ized by ou componen s, namely a uzzi ie , an in e ence engine, a knowledge base, and a de uzzi ie . The inpu s o a ypical uzzy sys em can be c isp da a (nume ic) and linguis ic alues ( uzzy se s). In he case o a c isp inpu , he uzzi ie assigns o i he applicable uzzy se and his p ocess is known as uzzi ica ion. Then, he in e ence engine accomplishes mapping o he inpu a iable alues o he linguis ic alues o he ou pu a iable h ough a sui able app oxima e easoning me hod wi h expe knowledge indica ed by a se o uzzy condi ional ules in he knowledge base. The knowledge base en ails he applica ion o domain knowledge which can be di ided in o a da abase and a ule base. The da abase comp ises linguis ic con ol ules, and he ule base includes domain expe knowledge. In addi ion o linguis ic alues, i nume ic da a ou pu is needed, hen de uzzi ica ion assigns c isp da a o he esul ing uzzy se . The classi ica ion o hea disease isk based on pa ien s’ heal h da a is pe o med using a uzzy in e ence sys em (FIS), and he algo i hm is p esen ed as Algo i hm 1. Elec onics 2022,11, 2292 7 o 19 Algo i hm 1: Classi ica ion o pa ien s’ heal h da a using FIS. S ep 1. The inpu s and he espec i e membe unc ions µ1de e mines he uzzy sys em S ep 2. Asce ain hea disease isk s a e using µ1(ECG1), µ1(MaxHea Ra e1), µ1(BloodP essu e1) as µ1(no mal) o µ1(low) o µ1(high) S ep 3. I Heal h isk s a e = µ1(high) 3.1 Send ale o GDusing SPARK as RTA 3.2 S o e Heal h isk s a e o he Puid in CS S ep 4. O he wise send Heal h isk s a e o he Puid o CS S ep 5. End he p ocess The inpu s o maximum hea a e, ECG, and blood p essu e, a e c ea ed and membe unc ion a e ed, which a e uzzi ied in o uzzy se s using a uzzy alue ange. Figu e 2 p esen s he wo king o FIS o hea disease isk p edic ion. Elec onics 2022, 11, x FOR PEER REVIEW 7 o 20 μ1 (BloodP essu e1) as μ1 (no mal) o μ1 (low) o μ1 (high) S ep 3. I Heal h isk s a e = μ1 (high) 3.1 Send ale o GD using SPARK as RTA 3.2 S o e Heal h isk s a e o he Puid in CS S ep 4. O he wise send Heal h isk s a e o he Puid o CS S ep 5. End he p ocess The inpu s o maximum hea a e, ECG, and blood p essu e, a e c ea ed and membe unc ion a e ed, which a e uzzi ied in o uzzy se s using a uzzy alue ange. Figu e 2 p esen s he wo king o FIS o hea disease isk p edic ion. Figu e 2. FIS o hea disease isk p edic ion. The uzzy se s hus c ea ed a e gi en as inpu o he FIS o classi ying pa ien s based on hei heal h da a. Table 2 depic s he linguis ic a iables and hei co espond- ing uzzy se o he FIS. Table 3 shows he membe unc ion and ange o he blood p es- su e a iable. Table 2. FIS—linguis ic a iable and uzzy se . Linguis ic Va iable Fuzzy Se Max Hea Ra e {Low isk, No mal, High isk} ECG {Low isk, No mal, High isk} Blood P essu e {Low isk, No mal, High isk} Table 3. Blood p essu e—membe unc ion and i s ange. Membe Func ion Range Low [40/90–70/100] No mal [70/110–80/120] High [90/130 and abo e] The inpu a iable alue is mapped in o he ou pu a iable’s linguis ic alues h ough a sui able app oxima e easoning me hod gi en as uzzy condi ional ules in he knowledge base. The esul s a e classi ied in unc ion o hese uzzy ules in he ule base along wi h co esponding membe unc ions. The no i ica ion is sen ega ding high- isk pa ien s, and he o e all pa ien isk s a us is s o ed in he cloud o u u e analysis. The da a o pa ien s classi ied as high isk o hea disease a e subjec ed o u he analysis in he ensuing p edic ion laye . 2.3. Da a P edic ion Laye Sequence p edic ion challenges ha e exis ed o a long ime and a e o en ega ded as one o he mos challenging p oblems in he da a science sec o o ackle. 2.3.1. RNN Figu e 2. FIS o hea disease isk p edic ion. The uzzy se s hus c ea ed a e gi en as inpu o he FIS o classi ying pa ien s based on hei heal h da a. Table 2depic s he linguis ic a iables and hei co esponding uzzy se o he FIS. Table 3shows he membe unc ion and ange o he blood p essu e a iable. Table 2. FIS—linguis ic a iable and uzzy se . Linguis ic Va iable Fuzzy Se Max Hea Ra e {Low isk, No mal, High isk} ECG {Low isk, No mal, High isk} Blood P essu e {Low isk, No mal, High isk} Table 3. Blood p essu e—membe unc ion and i s ange. Membe Func ion Range Low [40/90–70/100] No mal [70/110–80/120] High [90/130 and abo e] The inpu a iable alue is mapped in o he ou pu a iable’s linguis ic alues h ough a sui able app oxima e easoning me hod gi en as uzzy condi ional ules in he knowledge base. The esul s a e classi ied in unc ion o hese uzzy ules in he ule base along wi h co esponding membe unc ions. The no i ica ion is sen ega ding high- isk pa ien s, and he o e all pa ien isk s a us is s o ed in he cloud o u u e analysis. The da a o pa ien s classi ied as high isk o hea disease a e subjec ed o u he analysis in he ensuing p edic ion laye . 2.3. Da a P edic ion Laye Sequence p edic ion challenges ha e exis ed o a long ime and a e o en ega ded as one o he mos challenging p oblems in he da a science sec o o ackle. Elec onics 2022,11, 2292 8 o 19 2.3.1. RNN Deep lea ning algo i hms we e ex ensi ely esea ched and widely deployed in ecen yea s o ex ac ing in o ma ion om se e al ypes o da a. Neu al ne wo ks can lea n ep esen a ions and unco e p e iously unknown s uc u es. Nume ous deep lea ning a chi ec u es, namely he con en ional neu al ne wo k, deep neu al ne wo k, and ecu en neu al ne wo k ake in o accoun di e se aspec s o inpu da a [ 38 ]. In mos cases, CNN and DNN a e inep in coping wi h he inpu ’s empo al in o ma ion. RNNs p e ail in domains dealing wi h sequen ial inpu , such as ex , audio, o ideo. A cyclic connec ion is a common componen o he RNN design which allows he upda ing o i s cu en s a e depending on he cu en inpu and p e ious s a es [ 39 ]. The RNNs include he hidden o ecu en laye s, which consis o ecu en cells. The s a es o he ecu en cells a e impac ed by he cu en inpu ha has eedback connec ions and pas s a es. Di e en RNNs can be o med by o ganizing ecu en laye s in o di e en a chi ec u es. Thus, he ecu en cell, as well as he ne wo k a chi ec u e, dis inguish RNNs. The capabili y o RNNs is in luenced by a ying cells and hei inne connec ions. In some si ua ions, hese ne wo ks, such as comple e RNNs and selec i e RNNs, made up o con en ional ecu en uni s (sigma cells and anh cells), ha e showcased phenomenal success. Howe e , he RNNs wi h s anda d ecu en cells a e inadep a managing long- e m dependencies as i is daun ing o iden i y he in e connec ing in o ma ion wi h a conside able gap be ween he ela ed inpu da a. 2.3.2. LSTM Long sho - e m memo y (LSTM) has been p oposed o con end wi h “long- e m dependency” as he ou come o exhaus i e esea ch on RNNs, aimed a sequence lea ning. Long- ange in e dependence and nonlinea dynamics can be cap u ed using LSTMs [ 40 ] as i unc ions as he e ined e sion o RNN, wi h he hidden laye uni o he memo y cells in he place o ecu en uni s. Figu e 3illus a es he gene ic LSTM model. Elec onics 2022, 11, x FOR PEER REVIEW 9 o 20 whe e , i, o, and c a e he o ge ga e, inpu ga e, ou pu ga e, and cell ac i a ion ec o s, espec i ely, while W( , i, o, c) and b( , i, o, c) co espond o hei weigh ma ices and bias ec o s, espec i ely, and h ep esen s hidden alue. The e m 𝑥𝑡 e e s o he inpu o he memo y cell a ime while 𝑐𝑡 and 𝑐𝑡  deno e he cu en and p e ious memo y cell uni s. Figu e 3. Gene ic LSTM model. 2.3.3. P oposed Bi-LSTM Model The cons ain o he LSTM cell is ha i can ac on p io con en bu no on he u- u e one. Bidi ec ional ecu en neu al ne wo ks consis ing o wo dis inc LSTM hidden laye s wi h compa able ou pu in opposing di ec ions we e pu o h. P e ious and u- u e in o ma ion is used in he ou pu laye using his app oach. In Bi-LSTM, an inpu sequence 𝑋 = (𝑋1,𝑋2,…,𝑋𝑛) is compu ed in he o wa d di ec ion as ℎ𝑖 󰇍 󰇍 󰇍 = (ℎ1 󰇍 󰇍 󰇍 󰇍 ,ℎ2 󰇍 󰇍 󰇍 󰇍 ,⋯,ℎ𝑛 󰇍 󰇍 󰇍 󰇍 󰇍 󰇍 󰇍 ) and in he backwa d di ec ion as ℎ𝑡  󰇍 󰇍 󰇍 = (ℎ1  󰇍 󰇍 󰇍 󰇍 ,ℎ2  󰇍 󰇍 󰇍 󰇍 ,…,ℎ𝑛  󰇍 󰇍 󰇍 󰇍 ). The inal ou pu o his cell 𝑦𝑡 is c ea ed by bo h ℎ𝑖 󰇍 󰇍 󰇍 and ℎ𝑡  󰇍 󰇍 󰇍 , and he inal ou pu sequence is 𝑦 = (𝑦1,𝑦2,…𝑦𝑡…,𝑦𝑛). In deep ne wo ks, he chosen ac i a ion unc ion p o oundly in luences he aining dynamics along wi h he ask pe o mance. The ac i a ion unc ion p oposed by he Google B ain Team [41], Swish, s a ed as (x) = x.sigmoid(βx), was chosen o he p edic ion model. To add ess he cell di e gence issue o he gene ic model, a anh ac i a ion unc ion is included in he cell p opaga ion, and a leaky ec i ied linea uni (Leaky ReLU) is in- se ed a e ou pu ga ing. These collec i ely show he educed p edic ion oscilla ion and elimina ed nega i e ou pu s. Figu e 4 shows he LSTM cell s uc u e and Bi-LSTM o he p oposed model. 𝑓𝑡= swish(𝑊 𝑓𝑥𝑡+ 𝑊ℎ𝑓ℎ𝑡−1 + 𝑏𝑓) (7) 𝑖𝑡= swish(𝑊𝑖𝑥𝑡+ 𝑊ℎ𝑖ℎ𝑡−1 + 𝑏𝑖) (8) 𝑜𝑡= swish(𝑊 𝑜𝑥𝑡+ 𝑊ℎ𝑜ℎ𝑡−1 + 𝑏𝑜) (9) 𝑐𝑡  = anh(𝑊 𝑐𝑥𝑡+ 𝑊ℎ𝑐ℎ𝑡−1)+ 𝑏𝑐 (10) 𝑐𝑡= 𝑓𝑡∗ 𝑐𝑡−1 + 𝑖𝑡∗ 𝑐𝑡  + anh (11) ℎ𝑡= 𝑜𝑡∗ 𝑐𝑡 (12) 𝑦𝑡= 𝑜𝑡∗ 𝑐𝑡∗ Leaky ReLU (13) Figu e 3. Gene ic LSTM model. The memo y cells enable e aining and ou pu in o ma ion, he eby acili a ing he lea ning o long- e m empo al co ela ions. This includes sel -connec ions ha e ain he ne wo k empo al s a e and a e egula ed by h ee ga es: inpu ga e, ou pu ga e, and he o ge ga e. Ga ing is a p ocess ha de e mines he unc ion o each memo y cell in LSTMs. When he ga e is ac i a ed, he LSTM upda es i s cell s a e. The inpu and ou pu ga es go e n he low o memo y cell inpu s and ou pu s in o he emainde o he ne wo k. A o ge ga e was also in oduced o he memo y cell, which passes he high-weigh ed ou pu in o ma ion om one neu on o he nex . The in o ma ion e ained in memo y is de e mined by he inpu uni ’s high ac i a ion le el; i i is high, he memo y cell s o es he in o ma ion. Fu he mo e, a highly ac i a ed inpu uni will ans e in o ma ion o he ollowing neu on. Al e na i ely, high-weigh ed inpu da a a e s o ed in memo y cells. The Elec onics 2022,11, 2292 9 o 19 LSTM uni s’ ac i a ion is simila ly de e mined o RNNs. LSTM ne wo k in ol es mapping be ween inpu and ou pu sequence, i.e., X= (X1,X2, . . . , Xn)and Y= (Y1,Y2, . . . , Yn). =σW x +Wh h −1+b (1) i =σ(Wix +Whih −1+bi)(2) o =σ(Wox +Whoh −1+bo)(3) e c = anh(Wcx +Whch −1)+bc(4) c = ∗c −1+i ∗e c (5) h =o ∗ an h (c )(6) whe e ,i,o, and ca e he o ge ga e, inpu ga e, ou pu ga e, and cell ac i a ion ec o s, espec i ely, while W( ,i,o,c) and b( ,i,o,c) co espond o hei weigh ma ices and bias ec o s, espec i ely, and h ep esen s hidden alue. The e m x e e s o he inpu o he memo y cell a ime while c and e c deno e he cu en and p e ious memo y cell uni s. 2.3.3. P oposed Bi-LSTM Model The cons ain o he LSTM cell is ha i can ac on p io con en bu no on he u u e one. Bidi ec ional ecu en neu al ne wo ks consis ing o wo dis inc LSTM hidden laye s wi h compa able ou pu in opposing di ec ions we e pu o h. P e ious and u u e in o ma ion is used in he ou pu laye using his app oach. In Bi-LSTM, an inpu sequence X= (X1 , X2 , . . . , Xn) is compu ed in he o wa d di ec ion as → hi= ( → h1 , → h2 , · · · , → hn ) and in he backwa d di ec ion as ( h = ( ( h1 , ( h2 , · · · , ( hn ). The inal ou pu o his cell y is c ea ed by bo h → hiand ( h and he inal ou pu sequence is y= (y1,y2, . . . y . . . , yn). In deep ne wo ks, he chosen ac i a ion unc ion p o oundly in luences he aining dynamics along wi h he ask pe o mance. The ac i a ion unc ion p oposed by he Google B ain Team [ 41 ], Swish, s a ed as (x) = x.sigmoid( β x), was chosen o he p edic ion model. To add ess he cell di e gence issue o he gene ic model, a anh ac i a ion unc ion is included in he cell p opaga ion, and a leaky ec i ied linea uni (Leaky ReLU) is inse ed a e ou pu ga ing. These collec i ely show he educed p edic ion oscilla ion and elimina ed nega i e ou pu s. Figu e 4shows he LSTM cell s uc u e and Bi-LSTM o he p oposed model. =swishW x +Wh h −1+b (7) i =swish(Wix +Whih −1+bi)(8) o =swish(Wox +Whoh −1+bo)(9) e c = anh(Wcx +Whch −1)+bc(10) c = ∗c −1+i ∗e c + an h (11) h =o ∗c (12) y =o ∗c ∗Leaky ReLU (13) whe e ,i,o, and ca e he o ge ga e, inpu ga e, ou pu ga e, and cell ac i a ion ec o s, espec i ely, while W( ,i,o,c) and b( ,i,o,c) co espond o hei weigh ma ices and bias ec o s, espec i ely, and h ep esen s hidden alue. The e m x e e s o he inpu o he memo y cell a ime while c , e c , and y deno e he cu en and p e ious memo y cell uni s as well as he inal ou pu , espec i ely. Elec onics 2022,11, 2292 16 o 19 Elec onics 2022, 11, x FOR PEER REVIEW 17 o 20 Figu e 11. Compa ison wi h s a e-o - he-a sys ems. The esul s o he compa ison wi h he ela ed s a e-o - he-a hea disease p edic- i e sys ems e eal ha he p oposed sys em’s pe o mance su passes ha o he exis ing sys ems. Majo IoT-d i en asks o eal- ime sma sys ems in ol ing heal hca e wa an apid p ocessing as such applica ions a e delay and con ex -sensi i e. The escala ion in he numbe o IoT de ices and he upsu ge in he da a gene a ed by he sma de ices has esul ed in immense da a a ic esul ing in ex ensi e bandwid h u iliza ion and se ice di icul ies. As he Cloud–IoT model su e s om limi a ions such as la ency, connec i - i y, and bandwid h u iliza ion, he cloud compu ing model seems inadequa e o manage hese challenges solely due o i s cen alized model [43–47]. These sho comings se he s age o decen alized models o edge compu ing (EC) and og compu ing (FC), whe ein compu a ion and s o age can be handled a he edge nodes close o he da a sou ce. These newe compu ing echnologies complemen he Cloud and se e as an ex ension o i while enabling a i icial in elligence asks a he edge nodes. This hie a chical edge– og–cloud model conside ably educes he delay cons ain s by e icien ly handling he humongous da a acqui ed by he IoT de ices while mi iga ing la ency. Thus, he p o- posed cloud-based p edic ion sys em can be deployed a he og/edge laye s in he u u e o o e come he Cloud’s in insic cons ain s, such as inc eased la ency and bandwid h use, while managing he su ge in IoT da a. 80 84 88 92 96 100 Accu acy (%) La ha and Jee a Long e al. Mohan e al. Samuel e al. Ali e al. Paul e al. Ahmed e al. Kisho e and Jayan hi Dileep e al. Pam e al. Mehmood e al. Jabeen e al. Muzammal e al. Khan e al. Ali F e al. Zhang e al. P oposed Figu e 11. Compa ison wi h s a e-o - he-a sys ems. 7. Conclusions In his esea ch ini ia i e, an IoT–Cloud-based sma heal hca e sys em o hea dis- ease isk p edic ion is p oposed, and he uzzy in e ence sys em (FIS) and he ecu en neu al ne wo k’s bidi ec ional LSTM a e ha nessed o he p edic i e ask. The p oposed sys em’s accu acy, p ecision, sensi i i y, speci ici y, and F1-sco e a e 98.85%, 98.9%, 98.8%, 98.89%, and 98.85%, espec i ely, ou pe o ming o he s a e-o - he-a hea disease p edic- ion models. This is jus one ace o he heal hca e esea ch being done pe o med wi h p edic i e analy ics, wi h a huge po en ial o deep lea ning models ye o unco e . The model can be enhanced o au oma ically elici a pe sonalized die and exe cise ecommenda- ions o indi iduals as pe hei heal h condi ion and hea specialis ad ice. The p oposed sma hea disease p edic ion sys em u ilizes IoT de ices o da a acquisi ion, and o he p edominan asks a e ese ed o he Cloud. In he u u e, his wo k can be ex ended o include og/edge compu ing, whe ein ime-c i ical analy ical asks can be accomplished a he og/edge laye s o o e come he inhe en limi a ions o he Cloud, such as inc eased la ency and bandwid h u iliza ion while handling IoT da a upsu ge [ 48 – 51 ]. The e icacy o he heal hca e domain can be e olu ionized wi h p ecise and imely disease p edic ions alongside apid esponses and agile decision-making by clinicians, which will imp o e he o e all quali y-o -se ice when og/edge compu ing is in ol ed. Elec onics 2022,11, 2292 17 o 19 Au ho Con ibu ions: Concep ualiza ion, A.A.N., D.R., and K.S.; me hodology, A.A.N., D.R., and K.S.; so wa e, A.A.N.; alida ion, P.M.D.R.V., K.S., and D.G.R.; o mal analysis, A.A.N.; in es iga ion, A.A.N.; esou ces, K.S. and D.G.R.; da a cu a ion, A.A.N.; w i ing—o iginal d a p epa a ion, A.A.N.; w i ing— e iew and edi ing, A.A.N., D.R., P.M.D.R.V., K.S., and D.G.R.; isualiza ion, A.A.N. and K.S.; supe ision, D.R.; p ojec adminis a ion, D.G.R.; unding acquisi ion, D.G.R. All au ho s ha e ead and ag eed o he published e sion o he manusc ip . Funding: No unding was ecei ed o his s udy. Da a A ailabili y S a emen : The o iginal con ibu ions gene a ed o his s udy a e included in he a icle; u he inqui ies can be di ec ed o he co esponding au ho . Con lic s o In e es : The au ho s decla e no con lic o in e es . Abb e ia ions GDGa eway de ice CSCloud se e µ1Membe ship unc ion RTA Real- ime Analyze Puid Unique iden i ica ion numbe o pa ien Re e ences 1. Bha ia, M.; Sood, S.K. Game Theo e ic Decision Making in IoT-Assis ed Ac i i y Moni o ing o De ence Pe sonnel. Mul imed. Tools Appl. 2017,76, 21911–21935. [C ossRe ] 2. Fi ouzi, F.; Fa ahani, B.; Ma inšek, A. The Con e gence and In e play o Edge, Fog, And Cloud in he AI-D i en In e ne o Things (IoT). In . Sys . 2022,107, 101840. [C ossRe ] 3. Biswas, A.R.; Gia eda, R. IoT and Cloud Con e gence: Oppo uni ies and Challenges. In 2014 IEEE Wo ld Fo um on In e ne o Things (WF-IoT); IEEE: Manha an, NY, USA, 2014. 4. Bo a, A.; de Dona o, W.; Pe sico, V.; Pescapé, A. In eg a ion o Cloud Compu ing and In e ne o Things: A Su ey. Fu u e Gene . Compu . Sys . 2016,56, 684–700. [C ossRe ] 5. San os, G.L.; Takako Endo, P.; Fe ei a da Sil a Lisboa Tig e, M.F.; Fe ei a da Sil a, L.G.; Sadok, D.; Kelne , J.; Lynn, T. Analyzing he A ailabili y and Pe o mance o an E-Heal h Sys em In eg a ed wi h Edge, Fog and Cloud In as uc u es. J. Cloud Compu . Ad . Sys . Appl. 2018,7, 16. [C ossRe ] 6. Su esh, S. Big Da a and P edic i e Analy ics. Pedia . Clin. N. Am. 2016,63, 357–366. [C ossRe ] [PubMed] 7. Simpao, A.F.; Ahumada, L.M.; Gál ez, J.A.; Rehman, M.A. A Re iew o Analy ics and Clinical In o ma ics in Heal h Ca e. J. Med. Sys . 2014,38, 45. [C ossRe ] 8. Mio o, R.; Wang, F.; Wang, S.; Jiang, X.; Dudley, J.T. Deep Lea ning o Heal hca e: Re iew, Oppo uni ies and Challenges. B ie . Bioin o m. 2018,19, 1236–1246. [C ossRe ] [PubMed] 9. Pandey, S.; Janghel, R. Recen Deep Lea ning Techniques, Challenges and I s Applica ions o Medical Heal hca e Sys em: A Re iew. Neu al P ocess. Le . 2019,50, 1907–1935. [C ossRe ] 10. Muniasamy, A.; Tabassam, S.; Hussain, M.; Sul ana, H.; Muniasamy, V.; Bha naga , R. Deep Lea ning o P edic i e Analy ics in Heal hca e. In Ad ances in In elligen Sys ems and Compu ing; Sp inge : Cham, Swi ze land, 2019; pp. 32–42. [C ossRe ] 11. Smys, S. Su ey on accu acy o p edic i e big da a analy ics in heal hca e. J. In . Technol. Digi . Wo ld 2019 ,01, 77–86. [C ossRe ] 12. Amin, P.; Aniki eddypally, N.; Khu ana, S.; Vadakkemada hil, S.; Wu, W. Pe sonalized Heal h Moni o ing Using P edic i e Analy ics. In P oceedings o he 2019 IEEE Fi h In e na ional Con e ence on Big Da a Compu ing Se ice and Applica ions (BigDa aSe ice), Newa k, CA, USA, 4–9 Ap il 2019. 13. Joseph, P.; Leong, D.; McKee, M.; Anand, S.S.; Schwalm, J.-D.; Teo, K.; Men e, A.; Yusu , S. Reducing he Global Bu den o Ca dio ascula Disease, Pa 1: The Epidemiology and Risk Fac o s: The Epidemiology and Risk Fac o s. Ci c. Res. 2017 ,121, 677–694. [C ossRe ] 14. Fuchs, F.D.; Whel on, P.K. High Blood P essu e and Ca dio ascula Disease. Hype ension 2020 ,75, 285–292. [C ossRe ] [PubMed] 15. Sapp, P.A.; Riley, T.M.; Tindall, A.M.; Sulli an, V.K.; Johns on, E.A.; Pe e sen, K.S.; K is-E he on, P.M. Nu i ion and A he oscle- o ic Ca dio ascula Disease. In P esen Knowledge in Nu i ion; Else ie : Ams e dam, The Ne he lands, 2020; pp. 393–411. 16. Ca dio ascula Diseases. A ailable online: h ps://www.who.in /heal h- opics/ca dio ascula -diseases# ab= ab_1 (accessed on 14 June 2022). 17. Mo eno-Iba a, M.; Villuendas-Rey, Y.; Ly as, M.; Yáñez-Má quez, C.; Salgado-Ramí ez, J. Classi ica ion o Diseases Using Machine Lea ning Algo i hms: A Compa a i e S udy. Ma hema ics 2021,9, 1817. [C ossRe ] 18. La ha, C.; Jee a, S. Imp o ing he Accu acy o P edic ion o Hea Disease Risk Based on Ensemble Classi ica ion Techniques. In o m. Med. Unlocked 2019,16, 100203. [C ossRe ] Elec onics 2022,11, 2292 18 o 19 19. Long, N.; Meesad, P.; Unge , H. A Highly Accu a e Fi e ly Based Algo i hm o Hea Disease P edic ion. Expe Sys . Appl. 2015 , 42, 8221–8231. [C ossRe ] 20. Mohan, S.; Thi umalai, C.; S i as a a, G. E ec i e Hea Disease P edic ion Using Hyb id Machine Lea ning Techniques. IEEE Access 2019,7, 81542–81554. [C ossRe ] 21. Samuel, O.W.; Asogbon, G.M.; Sangaiah, A.K.; Fang, P.; Li, G. An In eg a ed Decision Suppo Sys em Based on ANN and Fuzzy_AHP o Hea Failu e Risk P edic ion. Expe Sys . Appl. 2017,68, 163–172. [C ossRe ] 22. Ali, L.; Rahman, A.; Khan, A.; Zhou, M.; Ja eed, A.; Khan, J.A. An Au oma ed Diagnos ic Sys em o Hea Disease P edic ion Based on χ2S a is ical Model and Op imally Con igu ed Deep Neu al Ne wo k. IEEE Access 2019,7, 34938–34945. [C ossRe ] 23. Paul, A.K.; Shill, P.C.; Rabin, M.R.I.; Mu ase, K. Adap i e Weigh ed Fuzzy Rule-Based Sys em o he Risk Le el Assessmen o Hea Disease. Appl. In ell. 2018,48, 1739–1756. [C ossRe ] 24. Ahmed, H.; Younis, E.M.G.; Hendawi, A.; Ali, A.A. Hea Disease Iden i ica ion om Pa ien s’ Social Pos s, Machine Lea ning Solu ion on Spa k. Fu u e Gene . Compu . Sys . 2020,111, 714–722. [C ossRe ] 25. Kisho e, A.H.N.; Jayan hi, V.E. Neu o-Fuzzy Based Medical Decision Suppo Sys em o Co ona y A e y Disease Diagnosis and Risk Le el P edic ion. J. Compu . Theo . Nanosci. 2018,15, 1027–1037. [C ossRe ] 26. Dileep, P.; Rao, K.N.; Bodapa i, P.; Goku uboyina, S.; Peddi, R.; G o e , A.; Shee al, A. An Au oma ic Hea Disease P edic ion Using Clus e -Based Bi-Di ec ional LSTM (C-BiLSTM) Algo i hm. Neu al Compu . Appl. 2022, 1–14. [C ossRe ] 27. Van Pham, H.; Son, L.H.; Tuan, L.M. A P oposal o Expe Sys em Using Deep Lea ning Neu al Ne wo ks and Fuzzy Rules o Diagnosing Hea Disease. In F on ie s in In elligen Compu ing: Theo y and Applica ions; Sp inge : Singapo e, 2020; pp. 189–198. [C ossRe ] 28. Mehmood, A.; Iqbal, M.; Mehmood, Z.; I aza, A.; Nawaz, M.; Nazi , T.; Masood, M. P edic ion o Hea Disease Using Deep Con olu ional Neu al Ne wo ks. A ab. J. Sci. Eng. 2021,46, 3409–3422. [C ossRe ] 29. Jabeen, F.; Maqsood, M.; Ghazan a , M.A.; Aadil, F.; Khan, S.; Khan, M.F.; Mehmood, I. An IoT Based E icien Hyb id Recommende Sys em o Ca dio ascula Disease. Pee Pee Ne w. Appl. 2019,12, 1263–1276. [C ossRe ] 30. Muzammal, M.; Tala , R.; Sodh o, A.H.; Pi bhulal, S. A Mul i-Senso Da a Fusion Enabled Ensemble App oach o Medical Da a om Body Senso Ne wo ks. In . Fusion 2020,53, 155–164. [C ossRe ] 31. Khan, M.A. An IoT F amewo k o Hea Disease P edic ion Based on MDCNN Classi ie . IEEE Access 2020 ,8, 34717–34727. [C ossRe ] 32. Ali, F.; El-Sappagh, S.; Islam, S.M.R.; Kwak, D.; Ali, A.; Im an, M.; Kwak, K.-S. A Sma Heal hca e Moni o ing Sys em o Hea Disease P edic ion Based on Ensemble Deep Lea ning and Fea u e Fusion. In . Fusion 2020,63, 208–222. [C ossRe ] 33. Zhang, D.; Chen, Y.; Chen, Y.; Ye, S.; Cai, W.; Jiang, J.; Xu, Y.; Zheng, G.; Chen, M. Hea Disease P edic ion Based on he Embedded Fea u e Selec ion Me hod and Deep Neu al Ne wo k. J. Heal hc. Eng. 2021,2021, 6260022. [C ossRe ] 34. Shukla, S.; Hassan, M.F.; Khan, M.K.; Jung, L.T.; Awang, A. An Analy ical Model o Minimize he La ency in Heal hca e In e ne -o -Things in Fog Compu ing En i onmen . PLoS ONE 2019,14, e0224934. [C ossRe ] 35. Kim, Y.; Bang, H. In oduc ion o Kalman Fil e and I s Applica ions. In In oduc ion and Implemen a ions o he Kalman Fil e ; In echOpen: London, UK, 2019. 36. Pa k, S.; Gil, M.-S.; Im, H.; Moon, Y.-S. Measu emen Noise Recommenda ion o E icien Kalman Fil e ing o e a La ge Amoun o Senso Da a. Senso s 2019,19, 1168. [C ossRe ] 37. Czabanski, R.; Jezewski, M.; Leski, J. In oduc ion o Fuzzy Sys ems. In Theo y and Applica ions o O de ed Fuzzy Numbe s; Sp inge In e na ional Publishing: Cham, Swi ze land, 2017; pp. 23–43. 38. Yu, Y.; Si, X.; Hu, C.; Zhang, J. A Re iew o Recu en Neu al Ne wo ks: LSTM Cells and Ne wo k A chi ec u es. Neu al Compu . 2019,31, 1235–1270. [C ossRe ] [PubMed] 39. Lip on, Z.C.; Kale, D.C.; Elkan, C.; We zel, R. Lea ning o Diagnose wi h LSTM Recu en Neu al Ne wo ks. a Xi 2015 , a Xi :1511.03677. 40. Hoch ei e , S.; Schmidhube , J. Long Sho -Te m Memo y. Neu al Compu . 1997,9, 1735–1780. [C ossRe ] [PubMed] 41. Ramachand an, P.; Zoph, B.; Le, Q.V. Sea ching o Ac i a ion Func ions. a Xi 2017, a Xi :1710.05941. 42. UCI Machine Lea ning Reposi o y. Uci.edu. A ailable online: h p://a chi e.ics.uci.edu/ml (accessed on 14 June 2022). 43. S ini asan, K.; Sha ma, A.; Anku , A. G oup Spa se Based Supe -Resolu ion o Magne ic Resonance Images o Supe io Lesion Diagnosis. In P oceedings o he 1s In e na ional Con e ence on Medical and Heal h In o ma ics, Taichung, Taiwan, 20–22 May 2017; ACM: New Yo k, NY, USA, 2017. 44. Mamdiwa , S.D.; Shak uwala, Z.; Chadha, U.; S ini asan, K.; Chang, C.-Y. Recen Ad ances on IoT-Assis ed Wea able Senso Sys ems o Heal hca e Moni o ing. Biosenso s 2021,11, 372. [C ossRe ] 45. S ini asan, K.; Gow haman, T.; Nema, A. Applica ion o S uc u al G oup Spa si y Reco e y Model o B ain MRI. In Ten h In e na ional Con e ence on Digi al Image P ocessing (ICDIP 2018), Shanghai, China, 11–14 May 2018; Jiang, X., Hwang, J.-N., Eds.; SPIE: Bellingham, DC, USA, 2018. 46. Jayalakshmi, M.; Ga g, L.; Maha ajan, K.; Jayakuma , K.; S ini asan, K.; Kashi Bashi , A.; Ramesh, K. Fuzzy Logic-Based Heal h Moni o ing Sys em o COVID’19 Pa ien s. Compu . Ma e . Con in. 2021,67, 2431–2447. [C ossRe ] 47. Ahsan, M.M.; Siddique, Z. Machine Lea ning-Based Hea Disease Diagnosis: A Sys ema ic Li e a u e Re iew. A i . In ell. Med. 2022,128, 102289. [C ossRe ] [PubMed] Elec onics 2022,11, 2292 19 o 19 48. Bha acha ya, D.; Sha ma, D.; Kim, W.; Ijaz, M.F.; Singh, P.K. Ensem-HAR: An Ensemble Deep Lea ning Model o Sma phone Senso -Based Human Ac i i y Recogni ion o Measu emen o Elde ly Heal h Moni o ing. Biosenso s 2022,12, 393. [C ossRe ] 49. P adhan, N.R.; Singh, A.P.; Ve ma, S.; Kau , N.; Roy, D.S.; Sha i, J.; Wozniak, M.; Ijaz, M.F. A No el Blockchain-Based Heal hca e Sys em Design and Pe o mance Benchma king on a Mul i-Hos ed Tes bed. Senso s 2022,22, 3449. [C ossRe ] 50. Vulli, A.; S ini asu, P.N.; Sashank, M.S.K.; Sha i, J.; Choi, J.; Ijaz, M.F. Fine-Tuned DenseNe -169 o B eas Cance Me as asis P edic ion Using Fas AI and 1-Cycle Policy. Senso s 2022,22, 2988. [C ossRe ] 51. Oyeleye, M.; Chen, T.; Ti a enko, S.; An oniou, G. A P edic i e Analysis o Hea Ra es Using Machine Lea ning Techniques. In . J. En i on. Res. Public Heal h 2022,19, 2417. [C ossRe ]