Meeting on the Management of Statistical Information Systems (MSIS 2013)

(Paris, France, and Bangkok, Thailand, 23-25 April 2013)


Topic (iii): Innovation


Improvement of data collection and dissemination by fuzzy logic


Invited/Supporting Paper


Prepared by Infostat with consultation in CEPS




Experimental tools based on fuzzy logic are designed with the objective of modernizing the collection and the dissemination of data in official statistics. In dissemination it is a tool capable of giving answers to imprecise questions in the search for relevant data.  It can solve more user demands and therefore improve image of NSIs as data provider.


In addition, tools based on the neural networks and fuzzy logic can help in the estimation of missing values by supervised and unsupervised searching for similar patterns.  Moreover, it can estimate attributes which are not in the focus of data collection but are relevant for national and international statistical institutes.


Furthermore, it can help identifying key customers (data users and respondents) and reveal their potential and weaknesses so as to ensure that similar customers/data providers are always similarly treated. The approach involves data mining of respondents’ behaviour and their classification into overlapping classes by flexible rules.


Modernization of the first and the last stage of data collection could create a chain reaction of improvements in data quality. Better data dissemination could motivate respondents to provide their own data timely and accurately and reduce the frequency of missing values implying more efficient imputation (less missing values and powerful soft computing tools). Finally, better and earlier data will be available for dissemination (websites) or exchange among institutes (e.g. by SDMX).


Theoretical and practical findings in Blue-ETS project (FP7) have revealed that all these goals could be reached by the soft computing approaches. Preliminary results are presented on small scale case studies on official statistics data. The next step is creation of a framework for the further development of full functional tools which could be applied in several parts of GSBPM.