Loading...

Comparison of Shuffling-Adaptive Neuro Fuzzy Inference System (Shuffling-ANFIS) with conventional ANFIS as feature selection methods for nonlinear systems

Jalali Heravi, M ; Sharif University of Technology | 2007

338 Viewed
  1. Type of Document: Article
  2. DOI: 10.1002/qsar.200630156
  3. Publisher: 2007
  4. Abstract:
  5. This paper compares the Shuffling-Adaptive Neuro Fuzzy Inference System (Shuffling-ANFIS) with conventional ANFIS as the method for selecting the features of nonlinear systems. Shuffling-ANFIS technique uses the advantage of data splitting with the ANFIS as a powerful feature selection method to select the most important factors affecting nonlinear phenomena. In this technique, the features with the largest percent of frequency can be found by running the conventional ANFIS sequential forward search on the large number of training and test sets using leave-one-out validation criteria. The superiority of Shuffling-ANFIS over the conventional ANFIS was evaluated by using two synthetic and one biological dataset. For all three sets, Shuffling-ANFIS method was superior in finding effective variables compared with five different conventional ANFIS and multiple linear regression technique. As a real dataset, a series of 80 1-[(2-Hydroxyethoxy)methyl]-6-(phenylthio)thymine (HEPT) derivatives acting as Non-Nucleoside Reverse Transcriptase Inhibitors (NNRTIs) was studied. Five descriptors were selected by Shuffling-ANFIS and were used to interpret the mechanism of HIV-1 inhibition of HEPT derivatives. A 5-5-1 network was developed using the five descriptors as its inputs. The accuracy of this network was illustrated by validation techniques of Leave-One-Out (LOO) and Leave-Multiple-Out (LMO) Cross-Validations (CV) and Y-randomization test. Superiority of Shuffling-ANFIS-ANN method with generalized bell membership functions over the previous works and also the five conventional ANFIS-ANN models indicates the power of this approach in selecting of the variables for nonlinear systems. © 2007 Wiley-VCH Verlag GmbH & Co. KGaA
  6. Keywords:
  7. 1 [(2 hydroethoxy)methyl] 6 (phenylthio)thymine ; RNA directed DNA polymerase inhibitor ; Thymine derivative ; unclassified drug ; Fuzzy logic ; Mathematical computing ; Priority journal ; Regression analysis ; Training ; Validation process ; Human immunodeficiency virus 1
  8. Source: QSAR and Combinatorial Science ; Volume 26, Issue 10 , 2007 , Pages 1046-1059 ; 1611020X (ISSN)
  9. URL: https://onlinelibrary.wiley.com/doi/10.1002/qsar.200630156