te Dota Analytics 1. Intesducten (pate BnalyBe- 1b is a process exteocting iseful “ine rmetion from tne raw deta, Zs Hi | i ava finaly tieh age | looks badeward | @ Analyticy is o futuce ae © Analysis with pest ox predicted wcesult - BR) Swpaile ‘In scaler focepe ® more exttnsible aller # Fools — Excel Nedext ete @ Tools — Ri pyth?r - sit is @_ process of using Stetistical technique to destribe set of dota. pse of deseibe the data = vadiablé—~ £15 a lobel we givete he dats: nn Froquanty af ceapeoted Forms - eg Histexical pie chart: items 19 graphical a the valdcs IO the ser. | @mean- THs swnply dhe avevage oh © median - the medians the celoulake o4 the | avewage eG the middle two date perbs - GSmade- Mode is jhe tnest frequently ecuring value In h | a@_doteset - @ Orspecsion — Sispessinn comet | Omange- diPference bet the lazzestand qroal lest value in _distAbution © yarienee— it is ® sede “The sndava deviation *s else _o earee rept ou ead cut the NUMbers awe + ee HS the square ef yaclance all chaps and size psod te indicate how spread outthe dete@ piagnesticuiAnalyter = ____________________> l ddvanced analytes that €semire dete Lcibis e . [ge Dansmer tha questions . “Why did it_ happen ] — used tow tnedetermine why Semedhing PePPened I ° [= ibs stetisHcal analy [measures of cleseviba tne strength £ divechenog | sis that ts Used aelatenship of two Variables: me ee i ee - 1b User the past data te create | these analytics abeut the und ex stending the Fur, a@_ model. sing the datd - @! - | - ib is the oven of business analytics dedicated te finding the best coucce of action few given situat © 2 ~ ib iso important aspects of any data analysis teesig, ~ toain purpese is t help look at data before making decision 7 Uses the undexstend tne Structi~e ot data} visualize —_©| muminsimbemigt— _ - itis a Und eestancting the exack Changes in weaeiab les that leads te changes, —_uhdexsfending the Use- Perception: [eee it is the process of 0 east meting the _s “elatehship omeng vaciables - = I —=—== ——————fe os 3

Pregrann - | output | Se = A _medel sepresents What wes Jearned by a machine learning algexithr - machine Jeaning medels pregrdr 15 comprised ef boty data ba procedure, for using ine date BIAS -— exer From incorrech assumption built $ inte model © voricinée — evvox from Sensitity in the cLeai ning Set- rosdel evalution - 15 a methed judge the correctness of mede/ on test data ae = Model evalutien i$ process taeeugh which we. quantity the Quality of a system's pxeclictions - Classification = confusion metre 7 Mean muerad Ervos a Reo mean Fptreen R-squarecl© (campuses unnerter x — | fp Confusion matrix shows the numb+r of covvect predictions rode b the Classification Land intervect aGiual outeomes TN the data tnode! Compared! te the [Daue pasive (TP) — occu When the med | predicts ance an observation balongs tea class andl in wea lity | it belongs te that class - i) Tede Negative scrw)- o6cur when the medel predichy that an obsexyation does nok beling too class and in meolity it dees not belongs tb thet class ) PaperpesiVeWFP)- occur when the medel predicts tnatan cbserwatien belongs te a class When in veolit. it dees not + Also known as typet ever. iw) Pale NeqaliveCeN) - OCcU*t Lohan dne model predicts that a9 observation belengste a class When in veality it does het . Also known as type F eter, predicted elass Negative [ Agtuai | pasitive | tp | close | Negative] Fp o| fecuwerey — “Th e accuwercy ot the Classifier is the-vatio of Cone ect peed tiene te the +teta) Number of Predictiens Tpe+tN REMACY = DP +ATN HRP HEN ol pwesicion — ——precisin= Tp fe specialty. Th oo Lae TPP. TN FPP _p | = ee ea bi TR+EN TN i TH FEN2 pinemoens — CReceiw es ~ operating characteristic) | SEEN plet that Sumevise tne performance - of biting classification _ ‘ [cue Positive ~ t ae oo ave curve — Caren cundas tne Roc cunwe). aaa = Meature frea undew the Cary eo =_pred = a Simple exomple_ «fF predichow mode! iS Pracchen ofthe selling price of a veal artete preper ty ued , = ect on thr ativibutey Clocatitn , square matty available Condifien et) a There ave three type of ecrer— ____O pesombeneees - = the absolute ervey cokulated — ervov calculated as ©. mean BeINeNeHFEIGIONEY ’ Mean angoluteene’ site knew as bt joasis one of the simples# Jose functen and an easy ate -understond —eyalulen Meti et: 7 nes eal ee!ib is alto knewnas ta loss. We calculstd tne ow by rguaving the diHteent 6 at? predicted values co | x Squared! Eero = Cyi) = rs ork is aleo called af Root Mean square Oeviatien fmSEz Jez ey Fie = jor15 one of ye most commen Teqresiiey loss Fun Chien ik Salto Knewnas Lo loss, we calculste he ecco by ayuering the dibbewent hat? predickd volver Astualy, aye SJuarec Exrov = Cyi Vi) Sf Dy Tb is aloo called ag Root mean sqyuave deviation ———_ ll Seeo 2, Machine Learning overview: = Etisa branch of computer science that aimy fp create Entelligant rnachineys - - tT is a humen being process natura] Intelligence, = human intelligense isa Quality that helps humans ty \eaen'ng Vind evstanding . and Seine Problerns, | Application — ° a Healthoare= healthcare Sacto 1S applying Az te take a better ¢ fastew Alagonosis + ® = AT 15 Used ‘mh cyberseaurity for _ Nehoork, porstecton: @ patermebiles- Az vsed te build saib-deiving vebicles = ALIS sotucity provide - __ Beducatiey — AL 1S Used automatic administrotiv< fasky ° Machine leaning - Jt = machine allow te learn frremy data ¢medel _ predictions based of experience * —— lone © Reqeessisry oF ID classttica tren {® clastetiny Fraud detechio a Speech wecgnyZatien - Frage receqnitatiin medical diagosis i a STS/SSO Translation / Cusbemer parfiling « ® Reummender Syctims. I® prediction - Teaphte predictien - soe© classifica ten - @ reqressien — = praqrany Jerk for pattern ‘1 ah unlabeled data = _ais— dataset patterns in datasets ~ it Faly bet? ug cso @ Reinforcement Learnings > WS 9 reward based Learning — Jearng by itera CHing with iby enyivenment Qype receives wewaval Per Comect # penalties 4, Incercect ‘in results - o Algeiinmt— G-leaming | pon = SUF dewing cows = Rebetiy 7 Garnnsing \ sunsupenieed ot Swain supenstd iy @ labelled daty \ Unlabetled data Pevkialy Jabeled i higher actuca @ | hiyzherer my ea lowes acereacy ewer een _ 2 We sma wx large daty - Mise lazgg daty dot set i \ Seb set @ tacks— ey, ification \ Cluskering Classi fica diay $ Meyeession | ee Chustering $9 || EHP) — Tt inv Filfog many decis | on different sampiy. of Same dataset beg _ __filgevithaa ~Randomy perestBoosting - Tt 's_a Sequential Precess Where cach sup- [sequent node\ attempt fe correct the evrew of Previous me det _ __Plgevithme— ca~ Boost 5 > _ib ‘involves fiking mang difterent madary typer one seme data $ Using ano: learn how to best ca medel to a} - 1S 4 generic termfor Mtatisticay metheds that attempt te fit a medel data =Rayrestien “invelves predicting _ test stoves Jabera trey, Molues jew prices ofan items cy. numeric data suchas —15 Used to predict +he Valve ofa vacable based of +he Value of anotner variable — De andent ¥oviable — va 4 bl< you want to predict 7 Zndependent varianie — Naviable yeu ore uring te predict ~ the ether variables value a Single independent varianle of numerical variable € than ene independent ures ice! varie Covev most of the date Panty sce teeOrcle~ prlynemicl q - Start with high odes $key oxdew pelynemia| - i p rreducan Tt As use when the. dependent varianles is Ses eate =Ysed for the binay classification preblenn J * | ipa | = ‘pra cess of recagnibiin, undursttonding? and grtuging | | of omyects ‘inte «elevant Jee ps 1@ prableme has only two oubceme © muitierass ~ penjem har mere tus outcomes * gers = © logistic we prestiin ® Support vette. machine @ ecisien teec. — d create the best line that con nadimentiinal spe |f inte classes. 1 | | 7 ——. ye = bulls a Sean ficay modes aie town spa Tee _ sheucieey arm ofa tree rou ping a Seto [19 the Some group oa tT PIES _O_ FAAS Segmentation @ prcument grouping = Ob mean 1 : maag Shift ees eecncec 1 © @ newjabeled dataset exe Complexe @ -type of unsypevised Lo @® unlabetico dataset. @ less campesc, {| | o | @ Data _pexforn well _on “smell te medium dataset @ teatures heed to be | manually identtied @ Hardwore 1s able to | funcHen of cpu B dataset is—usuailly ~rmertt to brodexcete tert ene, ® Terining time Guicle te tram | | © __perfermn wet! on larre / dlafasek « @ laams feature automati— : cally. @ Hardvore mequresiyni fr— cant Computing Power ey .cpy ® 4 time Computation nally Thbensivesocial Media and Text Analybler Ae * ie | | Inilions of peeple Usingsocial media platforms, —— hence, it required A veo] time monitering — __ 4 computing onc analysis of social media dete — it hetps tne hapbity of pesple — |— help csmmpanies bet bev Understand cthe nerd and expectation » oftheir Custemecs f- Im prov e the. efficiency of customer Sewice ae oll ° o. - (= The social mediq date analybics process begins with cleta_coptucing = [= It necessaxy +0 Coptuve valid ang welavant ata “inn pow tant- Land hence -this iS most | @ Sata understanding - Lp analyzing the captured date fex cetlectng oo meaningful ‘hfoxmaten (8 the. data yndexstending stages lies in the povddie. process. ¢ From the covet of social media onyly bier Lhpost ‘innper tent Stage “Io this entice prows) - pata presentation is the final Staze inthesecia| media analytics prececs — wesults ares presented mostly Using prope | date visualization -Ocrest - sucial media Text analy Laxteattion $ Analysis of textual element ens Lmedia content. Ex —tweeks- @nebsens — js sAnalysir con be employed te Erack | Follewexs te ielentify influential nedef fow thew position Inthe network - _ @ Petey — social media onalytics deal with extrac, _finalyzihg the Action perfornned by social 1 _Usexr Ge. Nike dislike, Shave. @ mesive - Mobile analytics deal with measuring ay. | Optimizing usex engagerment with Mobile -applicatie, © - Hyperlinks » A is about extractin: analy 2'ing $ Interpreting sectal media hyperlinks, @wesskien— is carried out +0 gain insight Prem the L geegraphic centent of Social media data, /___@ Seantbeengine — vp focuses on analysing Wistowen) Search data for gaining Valuable insight to range, © | Accessing -secial media data— ©, Using Apr's — mest ef Social media sites have then APL CApplication a Trtexfaces) 1@ Facbote — @® Witter i 6 Tnstageam oe ll j— Mest Websites clo not have, Apzts fox Stes slo ot have Apzs f En this case you have to use - vleb ~Seaping eMhith “towvelves Weitin __Ppregrarmte Fetch dota. frm van add Tent iets ate, Il ciccessing thei data Sicomputuce, 4 wees thd parse iSProgramming languazes like pytnen and rR ¢ Scientific Software likens offer packages fox interacting With APTS and have libraries | for ‘Woteeacting With most diyita) platfexme 3 | Fov- ex - —tTwiespy fox pytnen. and twitter for R | have became Standard e+ dewnleading tuittedate , ° ~ op [wine prediction— common features found in many Secial Networking sites for Possible Peiends Suggestion Found on Facaboste ox linkedin Garnrounity detection — share similar tesbex likes t and dislike « — expert Finding - finding solutims to wide variety of problems faad by pesple - = 18 a subset of CAL) - “Kespomssible fox +H. {| Undexstanding human language - S J — interacting beth human 4 computes. Application - @_pigita\ carts LO personal assistants ® peedictivetext @® SexSSee — brane Ine paragraph inte the Ts Separating of piece of deb inte Smaller un TE ma mt Hentences, — U8 Peer at frequency. if counts tne total ne of ently used bE 1b helps the tea ting Shortex clater of large a synenyrng - replace “the «even Word @_ Betton Keseacth Worle, @ The ier may me be uMetlich Maton @ Accuwace ® weule benec| |_cippreneh @ —Theword 1S valid word, _ _@ Accurae BS Enel Bi chichary bared appreach | Coxsespanding « that extrack tert fron a — dato . © Reduce Ceading time. ® selection process easier.EA I | friscsccectAReD URNS | Were en generating new sentences feemine iE owinal text + | — User in cheep leaning + = ® Nimmarication - 5 > Uses ‘in _bachiene learning __roriginal text extract anly pharef Fron the tot. 6 | oreand Analytics - a Coulecting ‘Information attempting to patter , — data Stoved dluving period of time Fidentify - © Tempral_teend Ahalytitr - deals with Hime. (2) Gee graphic trend Onalysis - location based Sanvices © || challenges to sotial mediq Prnralybicr— Qllvelume and velocity at cha Nange - @| Brexsity as challenge. @ || Posteuctura| party cut chatlen ga. @ || soci] media Analytics Accusacy.