From: Measuring the model risk-adjusted performance of machine learning algorithms in credit default prediction
Percentage of negative words over total words
Capital
3.87%
Credit scoring
3.09%
Provisions
4.88%