Deep Learning Techniques for Risk Modeling and Predictive Analytics

From the perspective of an academic, risk modeling represents the solution of an ill-posed inference problem where some risk level must be predicted with accuracy for a large population on the basis of limited historical data, possibly obscured by the presence of noisy predictors. Risk is generally understood in a probabilistic framework, and therefore risk models are often required to predict probabilities of default or default losses on the basis of sample means or sample proportions. Furthermore, risk modeling typically refers to the construction of predictive models for financial, operational, or insurance risks.