Implementing and Analyzing Fairness in Banking Credit Scoring Charlene Mariscal(a*), Fauzy Caesar Rochim(b), Yoga Yustiawan(b), Evawaty Tanuar(a)
(a) Bina Nusantara University [at]Bandung
Paskal Hyper Square, Jl. Pasir Kaliki No.25-27, Ciroyom, Kec. Andir, Kota Bandung, Jawa Barat
* charlene.mariscal[at]binus.ac.id
(b) Bank Rakyat Indonesia
Jl. Jend. Sudirman Kav. 44-46, Bendungan Hilir, Tanah Abang, Jakarta Pusat 10210
Abstract
The decision made by machine learning is mostly based on the historical data that used to train them. It raises the awareness that discrimination in machine learning should be eliminated, especially towards demographic features, such as gender or age, that may contain societal bias. Financial industry uses credit scoring as a reference to reflect the customer risk profile. To achieve fairness in the credit scoring model, this paper tries to: (1) assess bias in the model with different fairness metrics and (2) improve fairness in the ML model with several bias mitigation methods and algorithms. Moreover, the model performance also becomes a concern that should be preserved. This study depicts that some bias mitigation algorithms may work, but there is a trade-off with the performance. Implementing a reduction method performs the best to improve fairness and maintain the performance.
Keywords: Machine Learning- Bias and Fairness- Fair Machine Learning- Fairlearn- AI Fairness 360
