README.md

Ransomware-Transactions-Detection

Problem Statement:

Build a classification model to detect whether the bitcoin transaction is a ransomware attack or not.

Dataset:

The dataset is provided by UCI. It has 2,916,697 records with 10 features like the bitcoin transaction address and date. Each row represents a transaction. The target is the label of the transaction: Is it a ransom transaction or not?

Full data

It can be accessed from here: Full data

Best model data:

The data is too big to upload in github, but you can download it from here.

Model Score:

Data:

Undersampled (with TomekLinks and EditedNearestNeighbours) and unbalanced, scaled with min-max

Model:

AdaBoosting with XGB, GBT and GNB

Scoring:

Accuracy = 0.9118
Precision = 0.896
Recall = 0.9319
F-Score = 0.9136
AUC = 0.97

Recommendations:

For future work, we recommend collecting more updated data with more significant features like the time and targeted company information.