A neural network based on time series for spatiotemporal relationships prediction. (2016)

• Record Type: Journal Article
• Authors: Alouaoui, Hana; Turki, Sami Yassine; Faïz, Sami
• Is Part Of: International journal of spatial, temporal and multimedia information systems. Volume 1:Number 1(2016)
• Page Start: 63
• Page End: 86
• Publication Date: 2016
• Subjects: spatiotemporal data mining -- spatiotemporal association rules -- STAR -- spatiotemporal relationships -- relationship prediction -- neural networks -- nonlinear time series -- natural risk prediction -- spatial objects -- satellite images -- Tunisia; Geographic information systems -- Periodicals; Temporal databases -- Periodicals; Multimedia systems -- Periodicals; 005.753
• Languages: English
• ISSNs: 2052-3556

Abstract

The study of spatial objects and their evolution is a complex process. Most of the techniques used in this field do not consider the evolution of spatiotemporal relationships. The present paper proposes a new approach for the prediction of the future behaviour of spatiotemporal relationships based on spatiotemporal association rules. Such rules demonstrate the evolution of spatial objects and the influence of the spatial distribution of adjacent-areas relationships. We use a predictive neural network based on a nonlinear time series technique to generate spatiotemporal predictive rules. The learning examples correspond to spatiotemporal association rules and the results are in the form of spatiotemporal predictive rules assessing the future spatiotemporal relationships. These relationships can be used to inform about upcoming risks. We conduct an experimentation using a time series of satellite images, describing Megrine zone in the southern coast of Tunis (Tunisia). As a final result, we obtain spatiotemporal predictive rules describing the spatiotemporal relationships evolution between anterior and future dates. A comparison between the predicted values and the ground truth shows good correspondence rates varying between 78% and 90%.