9、数据建模与异常检测探索

数据建模与异常检测探索

1. 数据建模的困境

在对采购申请数据进行建模时，模型运行过程中显示了如下日志：

Train on 1450709 samples, validate on 483570 samples
Epoch 1/5
1450709/1450709 [==============================] 
- 19s 13us/step - loss: 8.4881e-04 - acc: 0.9999 -
val_loss: 0.0053 - val_acc: 0.9997
Epoch 2/5
1450709/1450709 [==============================] 
- 20s 14us/step - loss: 8.3528e-04 - acc: 0.9999 -
val_loss: 0.0062 - val_acc: 0.9997
Epoch 3/5
1450709/1450709 [==============================] 
- 19s 13us/step - loss: 8.5323e-04 - acc: 0.9999 -
val_loss: 0.0055 - val_acc: 0.9997
Epoch 4/5
1450709/1450709 [==============================] 
- 19s 13us/step - loss: 8.3805e-04 - acc: 0.9999 -
val_loss: 0.0054 - val_acc: 0.9997
Epoch 5/5
1450709/1450709 [===========