Artificial Intelligent Systems and Machine Learning

Suja S. Panicker ¹, P. S. Dhabe ², M. L. Dhore ³

Affiliations
1 Computer Engineering Department, MAEER's Maharashtra Institute of Technology, Pune, Maharashtra, IN
2 Vishwakarma Institute of Technology, Pune, Maharashtra, IN
3 Department of Computer Engineering, Vishwakarma Institute of Technology, Pune, Maharashtra, IN

Abstract

In this paper Fuzzy Min-Max Neural Network (FMN) classifier is used for Fault Diagnosis applications. It is a 3-layer architecture and uses a fuzzy membership function to reason about class label of a test pattern. We have collected two standard data sets-one from UCI repository and other from NASA, for experimentation purpose. Each data set is divided in two sets namely Training and Testing, using around half of the patterns. Above said Neural Network is trained using Training set and its performance is calculated using Test set. From the calculated performance it is found that the FMN performs well for both the data sets. By observing training, one can note that training time is more, but since training needs to be done only once it should not be treated as a serious handicap. Recall time per pattern is very small, thus the given neural network can be used for real time fault diagnostic purpose.

Keywords

Fault Diagnosis, Fuzzy Min Max Neural Network, NASA ADAPT Data, UCI Pump Data.

Full Text

     

P. S. Dhabe ¹, A. S. Natu ², A. S. Naval ², P. H. Parmar ², A. M. Padwal ², M. L. Dhore ²

Affiliations
1 Vishwakarma Institute of Technology, Pune, Maharashtra, IN
2 Department of Computer Engineering, Vishwakarma Institute of Technology, Pune, Maharashtra, IN

Abstract

In this paper fuzzy hyper line segment neural network (FHLSNN) is modified to represent the learned knowledge in a compact manner using a short hypothesis. This modification has increased percentage recognition rate and reduced recall time per pattern. We have also redefined the membership function so that it suit for the suggested modification. The performance of FHLSNN and modified fuzzy hyper line segment neural network (MFHLSNN) is tested and compared with the two splits of Heart disease database collected from UCI repository. The results shows that MFHLSNN algorithm is found superior to FHLSNN with respect to the recall time per pattern and space requirements for storing the hyper lines.

Keywords

Fuzzy Neural Networks, Heart Disease Detection, Medical Diagnosis, Pattern Recognition.

Full Text