Table 5.
Classification performance on predicting FDV “hands-off” offense types (best results are highlighted in green).
| Models | Resampling | Epochs | Accuracy | ROC | F1 | Precision | Setup Description |
|---|---|---|---|---|---|---|---|
| Naïve Bayes | - | - | 0.4673 | 0.4708 | 0.5091 | 0.5778 | - |
| MLP | - | 10 | 0.6034 | 0.5560 | 0.7359 | 0.6200 | 3 Dense; Dropout |
| - | 10 | 0.5642 | 0.5575 | 0.6499 | 0.6300 | 3 Dense; L1 Reg. | |
| 50% | 10 | 0.5673 | 0.5495 | 0.6665 | 0.6263 | 3 Dense; L1 Reg. | |
| 50% | 10 | 0.6067 | 0.5551 | 0.7545 | 0.6075 | 3 Dense; Dropout; L1 Reg. | |
| LSTM | - | 10 | 0.6036 | 0.5570 | 0.7369 | 0.6200 | 3 LSTM; Dropouts |
| 10% | 10 | 0.5928 | 0.5338 | 0.7298 | 0.6110 | 3 LSTM; Dropouts | |
| 50% | 10 | 0.5459 | 0.5333 | 0.6284 | 0.6244 | 3 LSTM | |
| 50% | 10 | 0.5939 | 0.5553 | 0.7150 | 0.6229 | 3 LSTM; Dropouts | |
| 50% | 10 | 0.5892 | 0.5532 | 0.7094 | 0.6217 | 3 LSTM; L1 Reg. | |
| Bi-LSTM | - | 10 | 0.5952 | 0.5632 | 0.7107 | 0.6300 | 3 Bi-LSTM; Dropouts; L1 Reg. |
| Bi-GRU | - | 10 | 0.6027 | 0.5641 | 0.7296 | 0.6200 | 3 Bi-GRU; Dropouts; L1 Reg. |
| BERT | - | 3 | 0.6072 | 0.5749 | 0.7367 | 0.6200 | MaxLen = 400; Batch size = 12 |