Please use this identifier to cite or link to this item:
|Title:||Performance evaluation of SAFIS for benchmark problems||Authors:||Sinara Vijayan.||Keywords:||DRNTU::Engineering::Electrical and electronic engineering::Control and instrumentation::Control engineering||Issue Date:||2008||Abstract:||This dissertation presents a performance evaluation of SAFIS algorithm with benchmark problems drawn from the areas of regression and classification. SAFIS is a Fuzzy Inference System recently developed by Rong Haijun based on the functional equivalence of a Fuzzy Inference System (FIS) and a Radial Basis Function Neural Network (RBFNN). After describing the SAFIS algorithm in detail, its performance evaluation based on two regression problems and two classification problems, is explained. The two regression problems considered here are Auto-mpg and California Housing. Proper selection of control parameters is highly essential to achieve best performance of the network. Here, a detailed sensitivity study of the SAFIS parameters is done and a study in selecting the range of the best algorithm parameters are done. Finally, performance evaluation of SAFIS with other algorithms is also done. Using two classification benchmark problems Image Segmentation and DNA, the selection of the optimal algorithm parameters is done and the performace of SAFIS for the classification problems is also studied with other algorithms. Based on these studies, an improved SAFIS algorithm called Fast SAFIS is developed. The optimal parameters of Fast SAFIS algorithm are determined and a performance evaluation of Fast SAFIS is done with SAFIS and other algorithms. The results indicate that the SAFIS algorithm exhibits very good performance with the regression benchmark problems in terms of training time, efficiency and number of hidden neurons to train the network. SAFIS performance is poor with the classification benchmark problems whereas Fast SAFIS performs very well with the same.||URI:||http://hdl.handle.net/10356/18787||Fulltext Permission:||restricted||Fulltext Availability:||With Fulltext|
|Appears in Collections:||EEE Theses|
Page view(s) 50243
Updated on Dec 5, 2020
Updated on Dec 5, 2020
Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.