An artificial neural network approach on catering premises inspection in Pahang state

• Main Authors: Hasan, Amran, Abu Bakar, Ibrahim, Md Ghani, Nor Azura, Ahmad Kamaruddin, Saadi, Miskon, Mohd Fuad
• Format: Article
• Language: English
• Published: Quality Scientific Publishing 2018
• Subjects: Q Science (General)
• Online Access: http://irep.iium.edu.my/63806/1/29581%20Published%20April%202018%20Amran.pdf; http://irep.iium.edu.my/63806/; http://www.journalcra.com

Background: The hygiene level of the premise reflect the safety and quality of the food served in the food services kitchen and the poor sanitary condition can contribute to food poisoning outbreaks. Recently, many food poisoning cases reported from food services sector and most of the cases are from institutional food services. These premises sometimes are graded as clean or very clean which can be questioned, mostly at institutions such as schools. Objective; The aim of this research is to identify the level of significance among the contributing factors which influence the caterers’ grading score in Pahang as the biggest state in Malaysian Peninsular using artificial neural network (ANN). Methods: In this research, the premises have been categorised into 3 categories namely Rest and Rescue Area (RnR) premises along the East Coast Highway, event caterers and institutional. A total of 268 premises were involved in this research with 66 (24.63%) RnR, 63 (23.51%) event caterers, and 139 (51.87%) institutional caterers. The instrument used in this research is based on the official risk based premise inspection form currently used by Ministry of Health Malaysia (MOH). The important items in the inspection form are process control, building and facilities, equipment and utensils, cleaning and maintenance, as well as food handler’s requirements. These items consist a total of thirty-one (31) elements with respected weightage score based on risk to food safety. The collected data is analysed using two-layer neural network with tansig-linear configurations, with trainlm activation function. Results: Prior to data normalization, the dataset is partitioned according 70-30-30 sets. In this research, the final model is reliable where the relative error of the training set is 0.076. The five most significant factors influencing the premises grades are critical control points (CCP), transportation condition, risky other related activity, adequate toilets, as well as adequate and safe water supply. Conclusion: As a conclusion, it is expected that the results will assist the related authorities to take appropriate actions prior to the important and compliance information, especially the significant aspects with respect to public health, permit, inspection and other related legal issues. It is suggested that the result can be improved by using other type of training functions such trainscg and trainbfg.