Table 3 Performance comparisons between the proposed method method (DL_\(\hbox {DA}_s\)) against the baselines, using stratified data augmentation (m: the augmentation data for each category is m as many as the original sample).

LR

0.685

0.628

0.848

0.800

0.550

0.833

0.626

0.601

0.830

LR_\({\hbox {DA}}_s\) (\(m=1)\)

0.671

0.563

0.850

0.840

0.569

0.866

0.586

0.521

0.832

LR_\({\hbox {DA}}_s\) (\(m=2)\)

0.678

0.589

0.861

0.820

0.556

0.897

0.606

0.557

0.841

SVM

0.745

0.716

0.895

0.840

0.582

0.908

0.697

0.692

0.880

SVM_\({\hbox {DA}}_s\) (\(m=1)\)

0.691

0.627

0.901

0.820

0.554

0.900

0.626

0.596

0.891

SVM_\({\hbox {DA}}_s\) (\(m=2)\)

0.725

0.670

0.903

0.820

0.556

0.900

0.677

0.651

0.901

MLP

0.745

0.721

0.898

0.820

0.562

0.896

0.707

0.703

0.886

MLP_\({\hbox {DA}}_s\) (\(m=1)\)

0.758

0.736

0.904

0.820

0.562

0.893

0.727

0.724

0.891

MLP_\({\hbox {DA}}_s\) (\(m=2)\)

0.758

0.739

0.910

0.820

0.562

0.905

0.727

0.726

0.896

LSTM

0.724

0.717

0.892

0.740

0.602

0.824

0.717

0.717

0.905

LSTM_\({\hbox {DA}}_s\) (\(m=1)\)

0.845

0.836

0.934

0.860

0.708

0.880

0.838

0.838

0.937

LSTM_\(\hbox {DA}_s\) (\(m=2)\)

0.818

0.805

0.952

0.880

0.763

0.963

0.788

0.786

0.940

CNN

0.772

0.757

0.912

0.860

0.589

0.881

0.727

0.728

0.902

DL_\({\hbox {DA}}_s\) (\(m=1)\)

0.772

0.751

0.922

0.860

0.590

0.861

0.727

0.726

0.910

DL_\({\hbox {DA}}_s\) (\(m=2)\)

0.785

0.763

0.926

0.840

0.577

0.868

0.758

0.754

0.917