Table 2 Performance measures for LSTM classifier for the non-transactional dataset
From: Modelling customers credit card behaviour using bidirectional LSTM neural networks
Description | Total | Missed Payments | Proportion | Accuracy (%) | Sensit-ivity (%) | Specif-icity (%) | AUC (%) | Brier Score (%) | KS | H-Measure |
|---|---|---|---|---|---|---|---|---|---|---|
All customers | 30,000 | 6636 | 22.12 | 82.4 | 37.51 | 95.15 | 78.47 | 13.28 | 0.43 | 0.3 |
Customers with at least one missed payment during the last 2 months | 15,265 | 3997 | 26.18 | 79.85 | 39.58 | 94.13 | 78.37 | 14.81 | 0.43 | 0.29 |
Customers with missed payment during last month | 13,714 | 3567 | 26.01 | 79.95 | 39.22 | 94.27 | 78.23 | 14.78 | 0.43 | 0.29 |
Customers with two consecutive missed payments | 7974 | 2592 | 32.51 | 77.33 | 51.04 | 89.99 | 79.81 | 16.15 | 0.46 | 0.32 |
Customers with three or more consecutive missed payments | 313 | 210 | 67.09 | 73.16 | 90 | 38.83 | 76.86 | 17.62 | 0.46 | 0.27 |