As illustrated before,the model's final output is a float number between [0,1].<br>
In previous tests, we assume there is fusion when the output is above 0.5 and get precision around 65%.<br>
To get lower false positive rate, we can increase the threshold of selecting fusions to 0.6/0.7/0.8/0.9.<br>
<br>
In the joint dataset .<br>
Accuracy on Test Set: 63.6296 %<br>
total: 9417  real=1,predict=1: 1863  real=1,predict=0: 2367  real=0,predict=1: 1058  real=0,predict=0: 4129<br>
Precision on Test Set: 63.7795 %<br>
Precision when above 0.6: 65.1246 %  total num (positive and false positive) is  843<br>
Precision when above 0.7: 74.1021 %  total num is  529<br>
Precision when above 0.8: 76.4463 %  total num is  242<br>
Precision when above 0.9: 83.8710 %  total num is  31<br>
<br>
We can see that,there is a clearly precision increase at higher threshold.