Applicaitons of reliability estimation and explanation of regression predictions in machine learning

In the field of decision support systems one can question reliability of automatically computed predictions, especially when dealing with risk-sensitive domains such as medicine, finance or industry. Wrong predictions of decision models can lead to costly professional or financial consequences. To provide insight into individual prediction reliability and understanding of decision models, we have developed two methodologies: (1) the methodology for estimating reliability of individual regression predictions, and (2) the methodology for explaining regression predictions and models. Both methodologies are independent of individual regression models used in the field of machine learning (eg, linear regression, regression trees and neural networks), enable providing trust into the decision support systems and expanding the expert knowledge about the domain. In the lecture, various applications of these methodologies will be presented.