Skip to content

Latest commit

 

History

History
26 lines (17 loc) · 1.83 KB

README.md

File metadata and controls

26 lines (17 loc) · 1.83 KB
Raw

Ponzhi Classfication

The folder ponzhi_scheme_detection contains all the files to generate the dataset and run the ponzhi_scheme_detection.

MAIN_DIR = ponzhi_scheme_detection/

DATA_DIR = ponzhi_scheme_detection/data/

TRANSACTIONS_DIR = ponzhi_scheme_detection/transactions/

Steps to run the code:

  1. cd ponzhi_scheme_detection
  2. create TRANSACTIONS_DIR if it doesn't exist with mkdir transactions. This folder will store all the transactions of the bitcoin addresses which are stored in the DATA_DIR
  3. The all_addresses.csv and ponzi_32.csv are taken from https://github.com/bitcoinponzi/BitcoinPonziTool/tree/master/CSV
  4. Run python merge_addresses.py located in MAIN_DIR to generate merged_addresses.csv in the DATA_DIR. The merged_addresses contains both the ponzi transactions and non ponzhi transaction addresses. We will use these addresses to get their respective transactions using the block explorer api
  5. cd data_collection
  6. Run python save_transactions.py to generate all the transactions of public addresses in json format in the TRANSACTIONS_DIR.
  7. Since 5. takes a lot of time, it is best to change MAX_DEGREE hyperparameter to dowload transaction details of addresses with less transactions first. I have downloaded all the addresses with transactions less than 25k.
    1. also generates a CSV in the DATA_DIR keeping track of files that have been succesfully downloaded. Using these transactions information, we will generate features to train Machine learning models.
  9. cd ../feature_generation and run get_features.py to generate features.csv in the DATA_DIR
  10. cd ../ and run the notebook Features_EDA_and_data_transform.ipynb to do EDA and generate transformed features
  11. run Classfication.ipynb to train models for the transformed features