A knowledge-driven service composition framework for wildfire prediction

Wildfire prediction has drawn a lot of researchers’ interest, but still presents a computational difficulty since it necessitates real-time data collected from several distributed data sources. Furthermore, because environmental Web services have, now, access to a wider range of environmental data sources, services might be functionally similar but of varying quality. In this paper, we propose a knowledge-driven framework for service composition that is based on a layered architecture. Based on these layers, the proposed framework aims to select the optimal service instances participating in a service composition schema, through a modular ontology to infer the quality of data sources (QoDS) and an outranking approach. Moreover, it aims to executing the service composition schema at runtime by dynamically readjusting both the service composition schema and the service instances via a machine learning-based service composition approach. The conducted experiments showed that the proposed framework enables (i) a reasonable reasoning time for assessing the data sources’ quality, (ii) a decrease in the ELECTRE III MCDM method’s execution time achieved by combining the skyline and α\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha $$\end{document}-dominance methods, (iii) dynamic generation of the most relevant service composition schema with the appropriate wildfire risk classes, and (iv) a high prediction accuracy using our proposed outranking approach compared to the randomly selected services.