This is a supervised machine learning model created by Surya Dheeshjith and Thejas Gubbi Sadashiva. The model is based on paper [1]. CFXGB is an extension of the model proposed in [2]. Implementation of cascaded forest is available at https://github.com/kingfengji/gcForest.
This repository also exists in this link with integrated Travis CI and Code coverage.
Implementation done in Python 3.7.3
For more details, contact Surya Dheeshjith : [email protected] (or) Thejas Gubbi Sadashiva : [email protected]
We have included demo code for execution and a detailed explanation of how you can use your own dataset with custom model parameters.
The implementation of using Parent nodes in the decision trees as stated in the paper has also been implemented
- All required packages are in requirements.txt
pip install -r requirements.txt
If facing issues downloading xgboost package, use this conda command
conda install py-xgboost
-
Datasets Used in [1] -
python3 Main.py -d Kad -p DefaultParameters.json
-
-h --help : List the parameters and their use.
-
-d --dataset : A dataset must be considered for learning. This parameter takes the dataset csv file name. This parameter must be passed.
-
-p --parameters : Model Parameters are passed using a json file. This parameter must be used to specify the name of json file. This parameter must be passed.
-
-i --ignore : Ignore the first column. (For some cases).
Default = False -
-r --randomsamp : Balance the dataset using random under sampling. (Use for imbalanced datasets).
Default = False -
-v --parentvaluecols [ BETA ]: Addition of columns based on class distributions of parents of leaf nodes in the decision tree.
Default = False -
-c --cores [ BETA ]: Number of cores to be used during addition of columns (When -v is True).
Default = -1 (All cores)
python3 Main.py -d <Dataset_Name> -p <Parameter_list>.json
[1] CFXGB: An Optimized and Effective Learning Approach for Click Fraud Detection.
[2] Z.-H. Zhou and J. Feng. Deep Forest: Towards an Alternative to Deep Neural Networks.