Modified anfis architecture with less computational complexities for classification problems
Adaptive Neuro Fuzzy Inference System (ANFIS) is one of those soft computing techniques that have solved the problems effectively in a wide variety of real-world applications. Even though it has been widely used, ANFIS architecture still has a drawback of computational complexities. The number of ru...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English English English |
| Published: |
2018
|
| Subjects: | |
| Online Access: | http://eprints.uthm.edu.my/275/1/24p%20NOUREEN%20TALPUR.pdf http://eprints.uthm.edu.my/275/2/NOUREEN%20TALPUR%20COPYRIGHT%20DECLARATION.pdf http://eprints.uthm.edu.my/275/3/NOUREEN%20TALPUR%20WATERMARK.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Adaptive Neuro Fuzzy Inference System (ANFIS) is one of those soft computing techniques that have solved the problems effectively in a wide variety of real-world applications. Even though it has been widely used, ANFIS architecture still has a drawback of computational complexities. The number of rules and its tunable parameters increase exponentially which created the problem of curse of dimensionality. Moreover, the standard architecture has a key drawback because of using grid partitioning and combination of gradient descent (GD) and least square estimation (LSE) which have problem to be likely trapped in local minima. Even though grid partitioning method is very useful to generate better accuracy for ANFIS model, since it generates maximum number of rules by considering all possibilities, but it also increases computational complexity. Since, ANFIS use fuzzy logic, the model accuracy is highly dependent on selecting the appropriate type of membership function. Furthermore, researchers have mainly used metaheuristic algorithms to avoid the problem of local minima in standard learning method. In this study, the experiments have been made to find out best suitable membership function for ANFIS model. Additionally, ANFIS architecture is modified for lessening computational complexities of the ANFIS architecture by reducing the fourth layer and reducing the trainable parameters as well. The proposed ANFIS model is trained by one of the metaheuristics approach instead of standard two pass learning algorithm. The performance of proposed modified ANFIS architecture is validated with the standard ANFIS architecture for solving classification problems. The results show that the proposed modified ANFIS architecture with gaussian membership function and Artificial Bee Colony (ABC) optimization algorithm, on average has achieved classification accuracy of 99.5% with 83% less computational complexity. |
|---|