Table 3.
Experimental results on NLST dataset (%, average (std) of cross-validation)
| Method | Accuracy | AUC | F1 | Recall | Precision | p-value |
|---|---|---|---|---|---|---|
| Ori CNN | 71.94(2.07) | 74.18(2.11) | 52.18(2.83) | 38.07(2.63) | 83.24(4.24) | <0.05 |
| MC-CNN | 73.26(3.10) | 77.96(0.98) | 59.39(3.70) | 47.91(4.87) | 78.62(3.09) | <0.05 |
| LSTM | 77.05(1.46) | 80.84(1.20) | 67.85(2.41) | 59.92(4.43) | 78.68(3.32) | <0.05 |
| Time-LSTM | 77.91(2.18) | 81.41(0.45) | 69.01(2.85) | 61.16(3.71) | 79.60(4.68) | <0.05 |
| tLSTM | 77.37(2.97) | 80.80(1.45) | 67.47(2.46) | 58.65(5.12) | 79.81(3.34) | <0.05 |
| DLSTM1 | 78.96(1.57) | 82.55(1.31) | 70.85(1.82) | 61.61(2.01) | 83.38(4.34) | *(base) |
| DLSTM2 | 78.63(1.45) | 81.51(1.11) | 68.35(2.03) | 57.49(3.87) | 84.88(4.56) | -- |
| DLSTM3 | 78.68(1.51) | 81.54(0.94) | 68.76(1.78) | 57.76(3.25) | 85.40(4.06) | -- |
| DLSTM4 | 78.05(2.01) | 82.09(1.38) | 68.90(2.52) | 59.84(3.48) | 81.44(3.43) | -- |
*
The average and standard deviation (std) of five-fold test results are reported.
*
The best average results are shown in bold. The p < 0.05 indicates our method ing significantly improve the compared method (McNemar test).
*
Methods citations: Ori CNN (Liao et al.. 2019), LSTM(Hochreiter and Schmidhuber, 1997) (Shi et al., 2015), Time-LSTM (Zhu et al., 2017), tLSTM(Santeramo et al., 2018).