Table 4: The experiment results for different LSTM networks.

Strategy ModelsAccuracy (%)AUCTPR (FPR = 0.1%)EER (%)Training times (h)

TAPLSTM ()71.430.886352.03-0.41
LSTM (α = 60)87.130.979181.676.450.54
LSTM (α = 80)91.560.985485.394.880.89
LSTM (α = 120)94.860.993191.533.17-

Truncated BPTTLSTM (α = 30)94.370.992891.113.10-
LSTM (α = 60)96.830.995093.37--
LSTM (α = 80)98.080.998995.13-1.24
LSTM (α = 120)98.470.999396.81-1.36
LSTM (α = 180)97.820.998795.42-1.53

Truncated BPTT + subsequence selectionLSTM (α = 120, = 90%)97.980.998895.011.341.16
LSTM (α = 120, = 95%)98.830.999797.220.84-
LSTM (α = 120, = 99%)98.660.999696.690.92-
LSTM (α = 120, = 97.5%)99.130.999998.690.541.34