HTML5 Webook

32/64

30Open InnovationBig Data Integration Research Center The Big Data Integration Research Center conducts research and development of data collection and analysis aimed at making eective use of real-space information obtained from the environment and human society. We are also de-veloping data mining technology that will integrate advanced environmental data with social data to analyze their cross-domain associations. This will facilitate model case studies of environmental influences on social systems such as transportation. We are also conducting R&D on methods that feedback the analysis results to sensors and devices in real space, and on sensor technologies that provide eicient and eective feedback. This will allow us to create, develop, and verify platform technology for implementing mechanisms capable of advanced situation recognition and behavior support, with the goal of optimizing social systems.R&D of fundamental technology for cross-domain data collabo-rationsWe are developing a data analytics plat-form for cross-domain data collaborations in order to exploit various forms of social big data and data acquired by various forms of sensing technology. We are also developing smart services for a smart sus-tainable society and constructing a data utilization testbed to support this develop-ment, and we are engaged in open inno-vation promotion activities involving re-gional demonstrations.So far, we have developed a method for discovering localized association patterns from heterogeneous urban sensing data through the mutual optimization of asso-ciation rule extraction and spatiotemporal clustering, and we have applied this meth-od to the prediction of traffic risks in the times of torrential rain based on rainfall data (XRAIN, phased array weather radar), traffic data (congestion, accidents), and social media data (Twitter). In an evalua-tion experiment, we achieved an improve-ment of about 60-80% in accuracy com-pared with a conventional method (Apriori) for association rule extraction alone, and we demonstrated the efficacy of our meth-od for association analysis of spatiotempo-rally-skewed sensing data such as data related to natural disasters. Also, to devel-op support for transportation and mobility, we prototyped an application system for risk-adaptive map-based navigation where routes are searched according to the us-er’s risk tolerance level based on traffic risks estimated from rainfall data. The re-sults of an evaluation performed using a drive simulator with 30 test users showed that approximately 86% of users select al-ternative routes to avoid risks when pre-sented with information about risk severity and driving costs (distance and time) (Fig.1).On the other hand, we are also con-ducting R&D on the use of atmospheric environment data. In collaborative re-search with Nagoya University, we have developed a portable sensor that acquires data on personal exposure to air pollution (PM2.5), and we have developed a method for predicting personal exposure based on lasso regression analysis. In preliminary tests conducted from March through May 2017, we confirmed that it is capable of pre-dicting personal exposure to PM2.5 with 80% accuracy (with an error range of ±15 µg/m3). We have also built a field demon-stration system on the NICT integrated Fig.1 : Mobility support based on a association analysis of heavy rain and trafc dataPrototype based on a commercial route guide service.Discovery of heavy rain xtrac data association ruleResearch and Development