Confidence Evaluation for Risk Prediction
This paper describe an application of a transductive method for risk mapping which allows to compute the confidence interval of an estimation, without any assumption on data distribution, except identity and independancy of inputs. The methos reliability is compared to conditionnal Sequential Gaussian Simulation. The robustness of this reliability to poor underlying regression models is also studied. The data set used is a digital elevation model of the South-West part of Switzerland ('Valais'). Experiments to evaluate the robustness of RRCM against the iid assumption are conducted using the data set of cadmium concentration in Leman Lake sediments in 1983 ('Cd83')
IDIAP-RR-01-22.pdf
openaccess
587.26 KB
Adobe PDF
92abf3a4e960942672669e246227dc25