Skip to main content

Table 5 Scorecard phase 1: “measuring the darkness of the algorithmic black-boxes”

From: Measuring the model risk-adjusted performance of machine learning algorithms in credit default prediction

   Lasso CART RF XGBoost MLP
Statistics Stability (SD) 0.001 0.005 0.003 0.001 0.002
Number hyper-parameters 0 1 2 2 + 5
Darkness Score 1 3 3 2 5
Technology Latency training 3.92 s 0.67 s 19.25 s 1.29 s 31.42 s
Transparency 1 2 3 3 5
Darkness Score 1 1 3 2 5
Market conduct Latency SHAP 0 0 60 s 8 s 2000 s
Darkness Score 1 1 2 2 3