Big data analytics through interval data methods

Abstract

Currently, the big data paradigm faces two colossal issues. The first one is to design hardware architectures and programming languages that allow efficient, reliable and relatively fast computation. Cloud computing, web services, distributed computation, Hadoop and Spark are some of the solutions offered until now. The second issue is to discover and develop methods of analysis for the huge and sometimes unstructured vast amount of information available in big data contexts, deciding which ones to use for real-time analytics. Machine learning and statistical analysis methods are two of the key pillars to be revisited in order to provide efficient solutions for the big data analytics (BDA) world. On an apparently different path, interval analysis (IA) has become an active research area since 1960, and an impressive development in the field of methods and applications of interval data has followed. However, the analysis, classification and forecasting of interval-valued data from the symbolic data analysis (SDA) approach is a very young research area, dating back less than 20 years, and still presents a wide array of open issues. This talk reviews methods of interval-valued data analysis in comparison with the corresponding crisp or single data methods from the BDA approach. In addition, it suggests some BDA contexts such as healthcare management, energy consumption or inflation rates; where the above approach can provide an efficient alternative way to process massive data and to get over some problems in classic BDA. Discussion of some open research questions is considered.