triageSVM.m

% function [resultsTable,labelCV] = triageSVM(inputX,Y)
    inputX = XY_pc(:,2:end);
    Y = XY_pc.(1);
    %Inistiates results tables; comment out if already created
    type = {};
    sensitivity = [];
    specificity = [];
    numObservations = [];
    numFeatures = [];
    errorRate = [];
    posPred = [];
    negPred = [];
    prevalence = [];
    genError = [];

    %Dummify X
    dumX = table;
    for c = 1:size(inputX,2)
        dumX.(c) = dummyvar(nominal(inputX.(c)));
    end
    
    %% Prep Data
    %Divide into train vs test data
    trainPrecentage = .7;
    trainCutoffIndex = fix(size(Y,1)*trainPrecentage);
    trainX = dumX(1:trainCutoffIndex,:);
    trainY = Y(1:trainCutoffIndex);
    testX = dumX((trainCutoffIndex+1):end,:);
    testY = Y((trainCutoffIndex+1):end);

    %Turn dumXs into matDumXs
    trainXMat = cell2mat(table2cell(trainX));
    testXMat = cell2mat(table2cell(testX));
    XMat = cell2mat(table2cell(dumX));
    
    

     %% Feature Selection
    c = cvpartition(Y,'k',5);
    opts = statset('display','iter');
    inmodel = sequentialfs(@my_fun,XMat,Y,'cv',c,'options',opts);
    
    %% Train & Test SVM
    %Train Model on TrainData, Predict with test input data & Analyze Performance
    SVMModel = fitcsvm(trainXMat,trainY,'Standardize',true,'KernelFunction','linear','ClassNames',{'SELF','DR'});
    % 'Crossval','on','KFold',10,'PredictorNames',{'Breathing Problem','BreathingRate','ChestPain','Cold/flu-severity','CoughOnset','CoughSpasms','CoughType','Cough-severity','CurrentState','Inhaled/Ingest','Meds/Respiratory','OtherSymptoms','PMH/RespiratoryDis.','Precipfactors?','Sorethroat-severity','Sputum','Temperature','TreatmentTried','WorstTime'}
    [labelTest,scoreTest] = predict(SVMModel,testXMat);
    [labelTrain,scoreTrain] = predict(SVMModel,trainXMat);
    CP_Test = classperf(testY, labelTest,'Positive',{'DR'}, 'Negative', {'SELF'});
    CP_Train = classperf(trainY, labelTrain,'Positive',{'DR'}, 'Negative', {'SELF'});

    %Train Model on AllData with Cross Validation & Analyze Performance
    SVMModelCV = fitcsvm(XMat,Y,'Crossval','on','KFold',10,'Standardize',true,'KernelFunction','linear','ClassNames',{'SELF','DR'});
    [labelCV,scoreCV] = kfoldPredict(SVMModelCV);
    CP_CV = classperf(Y, labelCV,'Positive',{'DR'}, 'Negative', {'SELF'});
    %Error Rate
    error = CP_CV.ErrorRate;
    %F Score
    R = CP_CV.Sensitivity;
    P = CP_CV.PositivePredictiveValue;
    fScore = 2*R*P/(R+P);
    %Generalization Error
    genError_CV = kfoldLoss(SVMModelCV);
    
    

    %% Write to Results Table
    type(end+1,1) = {'CP_Train'};
    numFeatures(end+1,1) = size(inputX,2);
    numObservations(end+1,1) = CP_Train.NumberOfObservations;
    sensitivity(end+1,1) = CP_Train.Sensitivity;
    specificity(end+1,1) = CP_Train.Specificity;
    errorRate(end+1,1) = CP_Train.ErrorRate;
    posPred(end+1,1) = CP_Train.PositivePredictiveValue;
    negPred(end+1,1) = CP_Train.NegativePredictiveValue;
    prevalence(end+1,1) = CP_Train.Prevalence;
    genError(end+1,1) = 0;

    type(end+1,1) = {'CP_Test'};
    numFeatures(end+1,1) = size(inputX,2);
    numObservations(end+1,1) = CP_Test.NumberOfObservations;
    sensitivity(end+1,1) = CP_Test.Sensitivity;
    specificity(end+1,1) = CP_Test.Specificity;
    errorRate(end+1,1) = CP_Test.ErrorRate;
    posPred(end+1,1) = CP_Test.PositivePredictiveValue;
    negPred(end+1,1) = CP_Test.NegativePredictiveValue;
    prevalence(end+1,1) = CP_Test.Prevalence;
    genError(end+1,1) = 0;

    type(end+1,1) = {'CP_CV'};
    numFeatures(end+1,1) = size(inputX,2);
    numObservations(end+1,1) = CP_CV.NumberOfObservations;
    sensitivity(end+1,1) = CP_CV.Sensitivity;
    specificity(end+1,1) = CP_CV.Specificity;
    errorRate(end+1,1) = CP_CV.ErrorRate;
    posPred(end+1,1) = CP_CV.PositivePredictiveValue;
    negPred(end+1,1) = CP_CV.NegativePredictiveValue;
    prevalence(end+1,1) = CP_CV.Prevalence;
    genError(end+1,1) = genError_CV;

    resultsTable = table(type,numFeatures,numObservations,sensitivity,specificity,errorRate,posPred,negPred,prevalence,genError);
end