Wyniki wyszukiwania - BazTech

1

Comparison of incomplete data handling techniques for neuro-fuzzy systems

Sikora M., Simiński K.

Computer Science

|

2014

|

Vol. 15 (4)

441--458

EN

Real-life data sets sometimes miss some values. The incomplete data needs specialized algorithms or preprocessing that allows the use of the algorithms for complete data. The paper presents a comparison of various techniques for handling incomplete data in the neuro-fuzzy system ANNBFIS. The crucial procedure in the creation of a fuzzy model for the neuro-fuzzy system is the partition of the input domain. The most popular approach (also used in the ANNBFIS) is clustering. The analyzed approaches for clustering incomplete data are: preprocessing (marginalization and imputation) and specialized clustering algorithms (PDS, IFCM, OCS, NPS). The objective of our research is the comparison of the preprocessing techniques and specialized clustering algorithms to find the the most-advantageous technique for handling incomplete data with a neuro-fuzzy system. This approach is also the indirect validation of clustering.

2

Neuro-rough-fuzzy approach for regression modelling from missing data

Simiński K.

International Journal of Applied Mathematics and Computer Science

|

2012

|

Vol. 22, no. 2

461-476

EN

Real life data sets often suffer from missing data. The neuro-rough-fuzzy systems proposed hitherto often cannot handle such situations. The paper presents a neuro-fuzzy system for data sets with missing values. The proposed solution is a complete neuro-fuzzy system. The system creates a rough fuzzy model from presented data (both full and with missing values) and is able to elaborate the answer for full and missing data examples. The paper also describes the dedicated clustering algorithm. The paper is accompanied by results of numerical experiments.

3

Merging of fuzzy models for neuro-fuzzy systems

Simiński K.

Theoretical and Applied Informatics

|

2011

|

Vol. 23, No. 2

107-126

EN

The merging of fuzzy model is widely used for reduction of rule number in fuzzy model. The supernumerosity of rules is mainly caused by grid partition of input domain. In the paper different cause for model merging is described. It is the need for creation of fuzzy model for large data set. In our solution the models are build basing data subset and then the submodels are merged into one. This approach enables quicker elaboration of submodels with relatively good knowledge generalisation ability without waiting for the whole data set to be processed. With passing time, the subsequent submodels are created and merged to create the better model.

PL

Artykuł opisuje scalanie modeli rozmytych w systemach neuronowo-rozmytych wykorzystywane przy tworzeniu modeli dla dużych zbiorów danych. Nieraz zbiory danych są tak duże, że nie jest możliwe wypracowanie modelu od razu dla całego zbioru. Tworzy się zatem modele dla podzbiorów zbioru danych. Uzyskane w ten sposób modele są następnie scalane, by wypracować jeden model. Podejście to jest także korzystne, gdy wszystkie dane nie są dostępne, ale są dostarczane partiami. Wtedy wstępny model jest wypracowany zanim wszystkie dane zostaną dostarczone do systemu. Artykuł przedstawia sposób wyznaczania podobieństwa reguł w modelu rozmytym oraz opisuje system neuronowo-rozmyty budujący i scalający modele wypracowane dla podzbiorów.