000			04979ntm a2200373 i 4500
999			_c99768 _d99774
003			MY-KuUP
005			20251125110754.0
006			t||||fr|||| 000 0
007			ta
008			230720t20232023my a|||fr|||| 000 0 eng d
020			_aTHE0009488 (Local) _qHardback
040			_aUMP _beng _cUMP _erda
090			_aKK .H36 2023 r Thesis
100	1		_aWang Hao, _eauthor.
245	1	0	_aArtificial intelligence-based customer requirement classification in quality management system using natural language processing for autonomous vehicles / _cWang Hao
264		1	_aPahang : _bUMP, _c2023
264		4	_c© 2023
300			_axvi, 128 pages : _bIllustration ; _c30 cm.+ _e1 CD ROM
336			_2rdacontent _atext
336			_2rdacontent _atext
337			_2rdamedia _aunmediated
337			_2rdamedia _acomputer
338			_2rdacarrier _avolume
338			_2rdacarrier _acomputer disc
347			_2rda _atext file _bPDF
500			_aCollege of Engineering
502			_aThesis (Doctor of Philosophy) -- Universiti Malaysia Pahang – 2023
504			_aIncludes bibliographical reference
520	3		_aMajor vehicle companies have created visible and significant outcomes and progress both in automotive and traffic systems. A quality management system could help vehicle companies identify customer requirements to fulfil them. Consumer demands are given in the format of large amount text data, classifying it becomes time, labour, and cost consuming. In addition, artificial intelligence (AI) techniques have many advantages, with the potential to replace humans. However, the popularity of autonomous vehicles faces not only technological difficulties, but also social barriers. Autonomous vehicle development needs a more efficient working process based on quality management systems. Most previous research on customer demands, requirements, and opinions are not comprehensive enough to provide deeper and more applicable data for car makers to develop and improve their products. Furthermore, there is no comprehensive corpus that is derived from a comprehensive customer requirements matrix. The conventional way to transfer customer requirements into usable text data is with low efficiency. Even though some Natural Language Processing (NLP) methods have been introduced into customer requirements classification, the performance of these methods have yet to be proven for a customised corpus in autonomous vehicle’s customer requirements classification. This study aims to build an efficient and simplified product development process based on quality management systems and embedded AI technology to create more comprehensive customer requirements that can be collected as input for autonomous vehicle developers. This will create a specific customer requirement corpus to compare the different conventional NLP algorithms based on a customised corpus and to compare the NaiveBayes, Maxent, and SVM Algorithms as well as improve the accuracy of classification by using a Convolution Neural Network (CNN). A three convolution layers with max pooling CNN is used in this study. Different quality management systems are compared, and a novel autonomous vehicle development process is created in this study. Then, Comprehensive customer requirements are collected by gathering former research outcomes based on different aspects. After that, A specific corpus based on both comprehensive customer requirements and NLP method is created. Then this corpus is expressed in the format of labelled data matrix and fed into different NLP algorithms to improve the efficiency of classification of customer requirements pertaining to autonomous vehicles. The training:testing dataset ratio set in 75:25, the datasets extended from 144 to 276 sentences, and the hyperparameters are optimised for CNN to get the best results. Meanwhile, for conventional classifiers use the common parameters. Average time is used for expressing the efficiency of the classification. Confusion matrix is used to evaluate the classification results for the conventional algorithms. Accuracy results is used for comparison. Results showed that the NLP method is applicable for autonomous vehicle development, and showed high efficiency in customer requirements’ classification, which could help reduce costs for car manufacturers. The classification accuracy using conventional classifiers reached lower than 50% for both training set ratio. The classification accuracy reached 90.48% under a 25% testing ratio and 100% under a 75% training ratio by using the CNN. In conclusion, even though this research is based on a customised corpus that cannot possibly compare with other research, the Convolution Neural Network showed its advantage in improving the accuracy of the classification when dealing with the specified corpus.
610	2	0	_aCollege of Engineering _xDissertations
650		0	_aUniversities and colleges _xDissertations
650		0	_aThesis
942			_2lcc _cTHESIS