Predicting the Probability of Exceeding Critical System Thresholds

• Дата: 2018
• Автори: Krammer, P., Kvassay, M., Hluchý, L.
• Формат: Стаття
• Мова: English
• Опубліковано: Інститут програмних систем НАН України 2018
• Назва видання: Проблеми програмування
• Теми: Моделі та засоби систем баз даних і знань
• Онлайн доступ: http://dspace.nbuv.gov.ua/handle/123456789/144599
• Назва журналу: Digital Library of Periodicals of National Academy of Sciences of Ukraine
• Цитувати: Predicting the Probability of Exceeding Critical System Thresholds / P. Krammer, M. Kvassay, L. Hluchý // Проблеми програмування. — 2018. — № 2-3. — С. 189-196. — Бібліогр.: 14 назв. — англ.

In this paper we show how regression modelling can be combined with a special kind of data transformation technique that improves model precision and produces several “preliminary” estimates of the target value. These preliminary estimates can be used for interval estimates of the target value as well as for predicting the probability that it has or will exceed arbitrary predefined thresholds. Our approach can be combined with various regression models and applied in many domains that need to estimate the probability of system malfunctions or other hazardous states brought about by system variables exceeding critical safety thresholds. We rigorously derive the formulas for the probability of crossing an upper bound and a lower bound both separately (one-sided intervals) and together (a two-sided interval), and verify the approach experimentally on a real dataset from the electric power industry.