Fraud classification is one of the most challenging problems in machine learning. Frauduland transactions are very minor event comparing to the legal ones. So, in Big data algorithms shows bias to the Majority events like non-frauduland transactions and results in high accuracy but low sensitivity.
In this project, we proposed a majority cluster based undersampling technique incorporating with random-forest classifier, which leverage the ensemble majority voting technique. This proposed model achieves, 95% accuracy with 99% sensitivity.
Install Julia Language: https://julialang.org/downloads/, Other necessary pakcages installation guide is given in the code.
Md. Tarikul Islam, Dipta Paul, Ishtiaq, Dewan Md. Farid