Authors:
George Marinos
1
;
2
;
Manos Karvounis
2
and
Ioannis N. Athanasiadis
1
Affiliations:
1
Data Competence Center, Wageningen University & Research, Wageningen, Netherlands
;
2
Agroknow IKE, Athens, Greece
Keyword(s):
Survival Analysis, Threshold Exceedance, Extreme Event Forecasting.
Abstract:
This study presents a novel framework designed for predicting threshold exceedance in time series data through the use of survival analysis techniques. Contrasting the traditional binary classification methodologies typically applied to this problem, our approach offers a unique perspective, modeling and predicting not only the occurrence but also the time-to-event information. This significant differentiation furnishes an invaluable tool for understanding and anticipating extreme events, and more importantly, it enhances decision-makers’ comprehension of temporal dynamics of such risks, enabling early intervention strategies. These facets are especially critical in various domains where timely action is essential. The effectiveness of our methodology has been empirically confirmed using both simulated and real-world datasets, showcasing our method’s precision in forecasting threshold exceedance. An illustrative application within the food safety domain, leveraging real-world data re
lated to food recalls over time, further demonstrates the practical utility of our approach, particularly in preventing and controlling high-risk hazards like salmonella. These findings underscore the wide-ranging implications of our method, particularly in applications where understanding the temporal dynamics of risks is paramount.
(More)