A survey on Machine Learning and Artificial Neural Networks

International Research Journal of Engineering and Technology (IRJET)

Volume: 09 Issue: 04 | Apr 2022

e ISSN: 2395 0056

p ISSN: 2395 0072

A survey on Machine Learning and Artificial Neural Networks

Shraddha Ovale

Omkar Nagalgawe

Mithilesh Rajput

1Assistant Professor, Dept. of Computer Engineering, PCCOE, Pune, Maharastra, India. 2,3,4,5Dept. of Computer Engineering, PCCOE, Pune, Maharastra, India.

***

Abstract - Artificial Intelligence is the branch of computer science that works on the different principles, methods, techniques used for complex problem solving and algorithms and study how the machines think and do things better than humans. Artificial intelligence works on knowledge, problem solving, planning, reasoning, thinking, communication, etc.

A survey paper on Machine Learning and Artificial Neural Networks is the overview of various techniques developed in Machine Learning and Artificial Neural Networks and their advantages, disadvantages, applications too. In short, we are adopting Machine Learning and Artificial Neural Networks in our day to day life to work more accurately, properly and efficiently, so the Artificial Intelligence industry is expected to grow more significantly over coming years. So, we must have prior knowledge of Machine Learning and Neural Networks, its techniques, algorithms and its applications and take the knowledge accordingly.

Key Words: ArtificialIntelligence,Machinelearning,Deep Learning,ArtificialNeuralNetworks.

1. INTRODUCTION

Artificial Intelligence is the most be known, vast and demanded field in computer science which deals with the ability of computer/machine to do things/works more precisely that are usually done by humans i.e., Biological Intelligence.

A boy can find the path but finding the path which goes homequicklyistheplacewherewerequireIntelligence.In such a way, the problem solving ability of humans and computersisthepointwherewemustworkonit.

TodayArtificialIntelligenceisintegratedinourday to day life in many forms such as assistants, robots, metaverse, gaming and many more. The various fields of Artificial Intelligence such as robotics, machine learning, deep learning,neuralnetworks,etc.arehelpsustoimprovethe efficiencyofworksinindustries,agriculture,healthcare,and education. So Artificial Intelligence is able to analyze language, handwritten text, visual effects like biometrics, facerecognition,speechrecognition,etc.

Machine learning is simply the study of computer algorithms, techniques which improve sufficiently, significantlyandmajorlythroughtheexperienceandlearn onitsown.Machinelearninghelpscomputerindeveloping models from data in order to automate decision making process. Algorithm used to imitate the way that human learns. Machine Learning is based on statistics, Linear Algebra, Probability and Calculus. Machine Learning Algorithmsworkonthebasisthatstrategies,algorithmsand interfacesthatworkedinpresentaswellasinfuturealso.

MachineLearninghastwoobjectives,

1)To classify data based on models which have been developed.

2)Tomakepredictionsforfutureoutcomesbasedonthese models.

DeeplearningisthepartofMachinelearningmethods.Deep refertouseofmultiplelayersinnetwork.ANNsalsousedin it for static and symbolic functions. This technique learns directlyfromdata andperformworks.Data canbeimage, text,fileorsound.

MachineLearningisfurtherclassifiedintothreetypesas showninfigure.

International

Volume: 09 Issue:

of Engineering and Technology (IRJET)

e ISSN: 2395 0056

p ISSN: 2395 0072

1)Supervised Learning: It hastraingdata i.e itconsist of input and output.The learning Algoriths such as Naives BayesAlgorithm works,analyzeit. Itis used whendata is labeledandithassometargetvariable.

2)Unsupervised: It has only input. The Input data is analyzed and output is generated. Most of the Machines Works on Unsupervised Learning. It is used when data is unlabeled and it does not have target variables. Uses K MeansClustering,Probabilisticclusteringmethods,Neural Networks,etc.

3)Semi Supervised: Itisacombinationofsupervisedand unsupervisedlearninganduseslabeledandunlabeleddata toimprovetheclassificationperformance.ItisusedinWeb Mining,TextMining,VideoMining,etc.Semi Supervisedable tosolve the problem which does not have enoughlabeled datatotrainsupervisedLearningAlgorithm.

4)Reinforcement: Itisanoptimaldecision makingprocess

TheagentsinteractwiththeenvironmentandWegotsome rewards, outputs which are based on behavior and gives criticalinformationaboutthealgorithm.Itconsistsoftwo processes namely Episodical Learning and Continuous Learning.

Artificial neural networks are used to produce artificial nerves which works same as biological nerves of humans and other animals. The collection of connected units or nodes called Artificial Neurons are the building blocks of Artificialneuralnetworksandittransfersmessagefromone partofmachinetootherandgivesoutputasaresultfrom thegiveninput.ArtificialNeuralNetworksalwayslearnby processing the examples. Each of these contains a known ‘Input’and‘Result’and this processingisdependingupon thedifferencebetweentheprocessedoutput(aprediction) and a target output. It is applicable in signal analysis, processing,monitoring,etc.

Asurvey paper on Machine LearningandArtificial Neural NetworksincludesabriefintroductionofMachineLearning and Neural Networks, various algorithms and concepts, advantages,disadvantages,anditsapplications.So,withthe help of this paper, you can be able to improve your knowledgeinArtificialIntelligence,MachineLearningand Artificial Neural Networks able to gain the knowledge of Machine Learning and Artificial Neural Networks can be usedinlargemannerinday to daylife.Wemustimprove thisfieldinlargeextentsothatweareabletoknowabout unknownmysteryoflifeanduniverse.

2. LITERATRE REVIEW/ RELATED WORKS

Miljanet.al,[1]proposedanapplicationofArtificialNeural Networks for hydrological modelling in karst various machineslearningalgorithmsofstate of the artforthetask of short term forecasting of river flow in a karst region describedinthispaperwithcomparativeanalysis.MLP,RBF NeuralNetworksandANFISwhichcontainneuralnetworks andfuzzy logicprincipleswithdifferent measuresinclude relative and absolute. On several locations, inputs of precipitationandflowdataanalyzed.Eightinputvariables usedinpreviouslytestedmodels,herefiveinputssubsetis used, For even better results. Five inputs subset model reduced complexity and the learning task has been accelerated. Support vector machine (SVM) used for regressionandclassificationanalysis.BunariverinBosnia andHerzegovinarelatedflowforecastonedayaheadbased ontheprecipitationandflowovertwopreviousdayswhich isoutperformedbytheANFISmodelandenabledabetter predictionmodel.

Wei Jin et.al, [2] proposed the basis of classification of machinelearningwithviewthatMachineLearningmainly focused supervised learning unsupervised learning and reinforcementlearningbutdoesnotfullyovercomeArtificial Intelligence.ThecommonalgorithmsinMachineLearning suchasrandomforest,boosting,ArtificialNeuralNetworks, decision tree, SVM, bagging, BP algorithms in proper manner. This helps us to improve the awareness about Machine Learning and increases the popularization of MachineLearninginday to daylife.

International Research Journal of Engineering and Technology (IRJET)

Volume: 09 Issue: 04 | Apr 2022 www.irjet.net

Mohsen Attaran and Promita Deb et.al, [3] proposed a model for successful and appropriate implementation of MachineLearninginorganization.Variousfactorsaddingthe evolution of Machine Learning. Improvement in data collection,analyticalsolutions,hardwareandsoftwareand computational power has been changed the field very drastically. Many jobs that require much training such as pattern recognition or image analysis which is done by pathologistsandradiologistsareindangerduetoMachine Learning.ThemainpointisthatcreatingaMachineLearning strategyinabusinessorganizationrequireahighlyqualified data scientists, organization is in favor of strategic investment because while adding new analytics functions affectexistingapplications,devices,servicesandwebsitesof thatorganization.

JI HAE KIM et.al, [4] proposed an Efficient Facial ExpressionRecognition(FER)Algorithmwhichdependson Hierarchical Deep Neural Network structure. Facial ExpressionRecognition(FER)istheinformationvisualform used to understand emotional situation of human. This research paper put forth an Efficient Facial Expression Recognition Algorithm by combining the appearance and geometric features by using Artificial Neural Networks to givemoreefficientandaccuratefacialrecognitionalgorithm. The co ordinate movement between Neural face and the peakemotionisextractedbythegeometricfeatures based networks.Theyconstructedstaticappearanceanddynamic feature combination from appearance network and geometric feature based networks. By experiments they showed that Top 2 error took place with average 82% correctnessusingtheappearancefeaturebasednetwork.An algorithm improved this error and achieved 96.5% correctness with 1.3% advancement. When comparing to otheralgorithmsintheCKTdataset.Theproposedalgorithm yields 91.3% of the correctness with 1.5% advancement whilecomparingwithotherexcitingmethodologiesinJAFFE dataset.

AlbertoRivaset.al,[5]proposedamethodforthedetection ofcattleusingdronesandConvolutionalNeuralNetworks.In this research paper convolutional neural networks (CNN) playsimportantroletoidentifycattlecapturedintheimage. Inthisarticle,adescriptionofinformationorknowledgeand itsperformanceinthedetectionofcattleisgiven.Thiscattle detection system has a high favorable outcome rate in identificationoflivestockfromimageviacamera.Currently, thisplatformhas87%accuracywhichisgoingtoincreaseby furtheralgorithmmodification.Thesatisfactoryaccuracyhas futurescopeofimprovement.Thesystemhasdrawbackof showinginaccuracywhensameanimalcrossesthepathof multicolor multi rotor several times. By increasing multi rotoranddividingareacansolvetheproblem.Ifthismodel hassuccessfulimpact,thentherewillbenoneedtoattach GPSdevicetoanimals.

Factor value:

e ISSN: 2395 0056

p ISSN: 2395 0072

R. Vargas et.al, [6] proposed that deep learning is an appearing region of Research in Machine Learning. It consists of number of hidden layers of Artificial Neural Networks. The recent achievement in Deep Learning architectureswithinmultiplefieldshavealreadyprovided notablecontributionsinArtificialIntelligence.Furthermore, thesuperiorbenefitsoftheDeepLearningmethodsandits hierarchynotonlyinlayersbutalsoinnon linearoperations are presented as well as compared with the more conventional Algorithms in the common applications. The most important value of Deep Learning depends on the optimization of earlier applications in Machine Learning. Due to its creativeness on hierarchical layers processing, DeepLearningprovidesuseffectiveoutputsindigitalimage processingandspeechrecognition. Today and in a future, Deep Learning results in a useful security tool due to the bothfacialrecognitionandspeechrecognitioncombined.

MaryamMNajafabadiet.al,[7]proposedthatDeepLearning hasabenefitofgivingasolutiontoaddressthedataanalysis and learning problems. Particularly, it assists in automaticallyuprootingcomplexdatarepresentationsfrom unsuperviseddata.ThismakesitavaluabledeviceforBig Data analytics, which consist of data analysis from huge collections of raw data. Data analytics tasks, mainly for analyzing huge data, information retrieval discriminative tasks such as classification and prediction. A survey of important literature research and application to different kinds of domains is explained in the paper as a means to showhowDeepLearningusedfordifferentpurposeinBig DataAnalytics.

OzerCelik,SerthanSalihAltunaydinet.al,[8]proposedthat MachineLearningisdependsupontheconceptoffindingthe best copy for new data among the previous data over increasingdata.Withtheadvancementsinthetechnologyin recenttime,Machinesplaysimportantroleinourdaytolife. Thedatainoursurrounding,inourlifeisincreasesdayby day.Thesedata areusedveryefficientlyandsignificantly. Firms that have recognized and invested in this area uses this technology frequently and achieving success. In such environment,theimportanceofInformationTechnologyand machines must be taken into consideration. Naive Bayes, SupportVectorMachine(SVM),K NN,Logistic regression, DecisionTreesaresomeoftheAlgorithmsandTechniquesin MachineLearning.

International Research Journal of Engineering and Technology (IRJET)

Volume: 09 Issue: 04 | Apr 2022 www.irjet.net

1. Hydrologic al modelling in Karst.

MLP, RBF NEURAL NETWORK ANDANFIS

1)The complexity decreasesas used ‘five inputs’ variables

2)The learningand accuracy increases

Additional things such as ground water level, geological composition of river basins, morphology and vegetation can influence prediction

5. Algorithm and methods for detection of cattle using Convolutio nal Neural Networks (CNN) and drones.

e ISSN: 2395 0056

p ISSN: 2395 0072

CNN 1)Accuracy of detection of cattle increases due to algorithms 2)There wouldbeno longer need to attach GPS device tocattle's

Problems of sameanimal crosspathof multirotor multiple times need tosolve

2. Classificati on of Machine Learning.

DECISION TREE,SVM, BP ALGORITH M

1)Various applications modulescan be design accordingto actualneeds ofusersand use.

2)Helps in the developmen t and advancemen ts in domestic enterprises

Machine Learning is supplementi ng not a traditional analytic method

6. Machine Learning consist of Artificial Neural Networks.

Convolutio nal Neural Networks (CNN), Supervised Semantics preserving Deep Hosting (SSPDH)

Feature Generation Automation, Better learning capabilities. Improves results and optimize processing time.

It consist multiple hidden layers of Artificial Neural Networks.

3. Conceptua lmodelfor Machine

DECISION TREES, SVM, NAIVE BAYS,KNN

1)Advancem ent in data collection.

2)Analytical solutions

Forvarious problems.

Implementa tion of Machine Learning in industries requires skilled data scientists anditiscost effective.

7. It gives a solutionto conveythe data analysis and learning processes.

Semantic Indexing, Super Vector Machine (SVH), Restricted Boltemann Machine (RBM)

Scalabity of Deep Learning Models, High dimensional ity. Can leverage unlabeled data.

Cannot easily learn with the linear and simpler models.

4. Facial Expression Recognitio n (FER) is used to understan d the emotional situation ofhuman

LBP FEATURE, CK+ AND JAFFE DATASET

1) With the help of algorithm accuracy of facial expression recognition can be enhanced

There is a scope to improve accuracy

8. Machine Learning depends upon the concept of findingthe best copy for new data among the previous data over increasing data.

K NN Education, image processing, Computatio nal Biology, Natural Language Processing

Requires large amount of data in order to perform work, expensive due to data complexity. Pruning occursmany times.

2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008

International Research Journal of Engineering and Technology (IRJET)

Volume: 09 Issue: 04 | Apr 2022 www.irjet.net

3. CONCLUSIONS

Machine Learning and Artificial Neural Networks uses AlgorithmsandtechniqueslikeSVM,CNN,K NNandBPfor different works.Thistechnique provesbeneficial butthey also have some drawbacks like Data Complexity, requires largeamountofdata,etc.InthisSurveywehavediscussed about applications of algorithms and architectures and challenges in front of us while working on Data. These techniques can be used in different fields in very efficient way like research and development. We require some techniques to find the solutions which are scalable, structured,etc.Advancedmodelscanhandledatacomplexity andbeabletouserule basedmodels.Infuture,researchwill be done in different domains in Machine Learning and ArtificialNeuralNetworks.

4. REFERENCES

[1] Miljan Kovačević, Nenad Ivanišević, Tina Dašić, Ljubo Marković. Application of artificial neural networks for hydrologicalmodellinginkarst.2016.

[2]WeiJin.ResearchonmachineLearninganditsAlgorithm and Development. Wei Jin 220 J. phys. conf.: ser. 1544012003.

[3]MohsenAttaran.MachineLearning:TheNew'BigThing' forCompetitiveAdvantages.January2018.

[4]JI HAEKIM,Byung GYUKim,ParthaPratimandDA MI Jeong. Efficient Facial Expression Recognition Algorithm Based on Hierarchical Deep Neural Network Structure. 10.1109/ACCESS.2019.2907327.

[5]Alberto Rivas,PabloChamoso,Alfonso BrionesandJuan Manuel Corchado. Detection of cattle Using Drones and ConvolutionalNeuralNetworks.

[6]R.Vargas,A.Mosavi,L.Ruiz.DeepLearning:AReview. AdvancesinIntelligenceandComputing.(2017)

[7] Maryam M Najafabadi, Flavio Villanustre, Taghi M Khoshgoftaar, Naeem Seliya, Randall Wald and Edin Muharemagic.Deeplearningapplicationsandchallengesin BigDataanalytics.(2015)2:1.

[8]OzerCelik,SerthanAltunaydin.AResearchonMachine LearningandItsApplications.(2018).

2022,

e ISSN: 2395 0056

p ISSN: 2395 0072

Impact Factor value: 7.529

9001:2008