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

Alonso Robisco, Andrés; Carbó Martínez, José Manuel

doi:10.1186/s40854-022-00366-1

Financial Innovation

Table 11 Terminology for each model risk category

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

Statistics	Technology	Market conduct
Stability	Transparency	Privacy
Over-fitting	Carbon footprint	Auditability
Over-fit	External providers	Interpretability
Hyper parameters	Dependencies	Biases
Dynamic calibration	Cyber-risk	Expert judgement
Feature engineering	Third-party	Expert personnel
Forecast	Cloud	Human judgement
Parameters	IT system	Human-in-the-loop
Test	Legacy	Human-on-the-loop
Calibration	Infrastructure	Governance
Features	Computing	Ethics
Explanatory variables	Computational power	Ethical
AUC	Deployment	Human
ROC	in-house	Compliance
Recall	Development	Management
Prediction	Pilot	Explainability
Logit	ICT	Internal control
Algorithm	Architecture	Knowledge
Scenario	Resilience	Consumers
Data quality	Security	Consumer protection
Back-testing	Operational risk	Discrimination
Benchmarking	Outsourcing risk	Uncertainty
Model	Reversibility	Accountability
Optimisation	DevOps	Market abuse
Dimensionality	Software	Complexity
Validation	Hosting	Decision making
Metrics	Cyber-risk	Soundness
Structured	Model stealing	Conduct
Unstructured	Poisoning attacks	Internal audit
Semi-structured	Adversarial attacks	Fairness
Classification	Open-source	Fair
Tree-based		Diversity
Neural network		Oversight
Regression		Simplicity
Clustering		GDPR
Support vector machine		Transparency
Reinforcement learning		Traceability
Parametric		Opaqueness
Non-parametric		Black-box
Performance		Black boxes
Prociclality		Surrogates
Train		Trust
Training		Trustworthiness
Volatility		Influence
Tuning		SHAP
Threshold		Shapley
Cross-validation		Independent conditional expectations
Compilation		Partial dependence plots
Out-of-sample statistical		ICE
Predictive		PDP
Challenger model		Complex
Correlation
Confidence level

Back to article page