Using convolution neural networks for improving customer requirements classification performance of autonomous vehicle

Customer requirements are vital information prior to the early stage of autonomous vehicle (AV) development processes. In the development process of AV many decisions have been made concerning customer requirements at the first stage. The development of AV that meets customer requirements will incre...

Full description

Bibliographic Details
Main Authors: Hao, Wang, Asrul, Adam, Fengrong, Han
Format: Article
Language:English
Published: University of Bahrain
Subjects:
Online Access:http://umpir.ump.edu.my/id/eprint/35173/1/IJCDS_120121_1570767667.pdf
Description
Summary:Customer requirements are vital information prior to the early stage of autonomous vehicle (AV) development processes. In the development process of AV many decisions have been made concerning customer requirements at the first stage. The development of AV that meets customer requirements will increase the global consumer and remain competitive. Safety and regulation are one of crucial aspect for customers that requires to be concerned and evaluated at the early stage of AV development. If safety and regulation related requirements did not well identified, AV developer could not develop the safest vehicles due to the huge compensation of accidents. To efficiently classify customer requirements, this study proposed an approach based on natural language processing method. For classification purpose, the customer requirements are divided into six categories that the concept are come from the quality management system (QMS) standard. These categories will be as input for the next process development in making the best decision. Most of conventional algorithms, such as, Naive Bayes, MAXENT, and support vector machine (SVM), only use limited human engineered features and their accuracy for customized corpus in sentences classification are proven low which is less than 50 percent. However, in literature, convolution neural networks (CNN) have been described efficiently to overcome the customized corpus of sentence classification problems. Therefore, this study implements CNN architecture in customized corpus classification operations. As the results, the accuracy of CNN classification has improved at least 6 percent compared to the conventional algorithms.