MachineLearning-PoweredDiagnosisofCoralReefDiseaseFactorsUsing PCAandFF-ANNTechniques

Abstract

Coral reefs representoneof themostvitalmarineecosystems, supportingawiderangeof
aquaticlife.However, risingoceantemperaturesandenvironmentalfluctuationsincreasingly
threatentheirsurvival.Thesechangesnotonlydamagethecoralsthemselvesbutalsodisrupt
thenumerousorganismsthatdependonthem.Toensureeffectiveconservation,itiscrucialto
identify and analyse the factors responsible for coral diseases. This study focuses on
predictingthecausalagentsofcoralreefdiseasesusingadata-drivenapproach.Theproposed
model investigatestwotypesofdatasets:metagenomicsequencedataassociatedwithlesions
ofCoralPatch(CP)andBlackBandDisease(BBD),andmetatranscriptomicdatacapturing
thebiologicalactivitywithinCPandBBDlesions.Bothdatasetscontainvaryingoccurrences
of the twodisease conditions, enabling a comprehensive examinationof the underlying
microbialcontributors.Twocomputationaltechniques—PrincipalComponentAnalysis(PCA)
and Feed-Forward Artificial Neural Networks (FF-ANN)—are employed to enhance
prediction performance. PCA is used for dimensionality reduction, extracting themost
informative features, while the FF-ANN utilizes a multilayer perceptron with
backpropagation toclassify theorganisms responsible for coral infectionsand identify the
most severedisease impacts.Experimental resultsdemonstrate that theproposedPCA–FF
ANNframeworkoutperformsconventionalSVMandCNNmodels,offeringamoreeffective
solutionfordiagnosingcoralreefdiseasecausality.