Here we have collected the code to accompany the preprint submitted to Decision Support Systems "Efficient Fraud Detection using Deep Boosting Decision Tree"
Folders:
-
"Data": This folder contains some pny-size data for reproduce the fraud detection data experiments.
-
"Script": This folder contains the code for deep boosting decision tree:
-
First, run the "DBDT-SGD" to run the baseline of deep boosting decision tree for balanced data.
-
To run the "DBDT-AUC" to train the deep boosting decision tree only by AUC maximization, but it can not get fine result.
-
To run the "DBDT-Com" to compositional train the deep boosting decision tree by AUC maximization and exponential minimization for fraud detection.
-
If you run all the experiments on multilayers with all the choices of training deep boosting decision tree as in our paper, "DBDT-Com-multilayers" will reproduce the results in the paper.
-