UUM Electronic Theses and Dissertation
UUM ETD | Universiti Utara Malaysian Electronic Theses and Dissertation
FAQs | Feedback | Search Tips | Sitemap

Self learning neuro-fuzzy modeling using hybrid genetic probabilistic approach for engine air/fuel ratio prediction

Al-Himyari, Bayadir Abbas (2017) Self learning neuro-fuzzy modeling using hybrid genetic probabilistic approach for engine air/fuel ratio prediction. Doctoral thesis, Universiti Utara Malaysia.

[thumbnail of s93024_01.pdf]
Preview
Text
s93024_01.pdf

Download (3MB) | Preview
[thumbnail of s93024_02.pdf]
Preview
Text
s93024_02.pdf

Download (946kB) | Preview

Abstract

Machine Learning is concerned in constructing models which can learn and make predictions based on data. Rule extraction from real world data that are usually tainted with noise, ambiguity, and uncertainty, automatically requires feature selection. Neuro-Fuzzy system (NFS) which is known with its prediction performance has the difficulty in determining the proper number of rules and the number of membership functions for each rule. An enhanced hybrid Genetic Algorithm based Fuzzy Bayesian
classifier (GA-FBC) was proposed to help the NFS in the rule extraction. Feature selection was performed in the rule level overcoming the problems of the FBC which depends on the frequency of the features leading to ignore the patterns of small classes. As dealing with a real world problem such as the Air/Fuel Ratio (AFR) prediction, a multi-objective problem is adopted. The GA-FBC uses mutual information entropy, which considers the relevance between feature attributes and class attributes. A fitness function is proposed to deal with multi-objective problem without weight using a new composition method. The model was compared to other learning algorithms for NFS such as Fuzzy c-means (FCM) and grid partition algorithm. Predictive accuracy and the complexity of the Fuzzy Rule Base System (FRBS) including number of rules and number of terms in each rule were taken as terms of evaluation. It was also compared to the original GA-FBC depending on the
frequency not on Mutual Information (MI). Experimental results using Air/Fuel Ratio
(AFR) data sets show that the new model participates in decreasing the average number of attributes in the rule and sometimes in increasing the average performance compared to other models. This work facilitates in achieving a self-generating FRBS from real data. The GA-FBC can be used as a new direction in machine learning research. This research contributes in controlling automobile emissions in helping the
reduction of one of the most causes of pollution to produce greener environment.

Item Type: Thesis (Doctoral)
Supervisor : Yasin, Azman and Horizon, Gitano
Item ID: 6809
Uncontrolled Keywords: Genetic Algorithms, Fuzzy Bayesian classifier, Rule extraction, Feature selection, Mutual Information Entropy.
Subjects: Q Science > QA Mathematics > QA76 Computer software > QA76.76 Fuzzy System.
Divisions: Awang Had Salleh Graduate School of Arts & Sciences
Date Deposited: 19 Sep 2018 04:01
Last Modified: 02 May 2021 00:51
Department: Awang Had Salleh Graduate School of Arts and Sciences
Name: Yasin, Azman and Horizon, Gitano
URI: https://etd.uum.edu.my/id/eprint/6809

Actions (login required)

View Item
View Item