Threat analysis using artificial neural network

Abstract

The purpose of this study is to explore the use of an Artificial Neural Network threat analysis tools for analyzing threats in healthcare system. The research method used a feed forward neural network which consisted of 50 input variables and one output. The datasets used in neural network are provided by previous research conducted in one of the Government Supported Hospital. The neural network is trained with the datasets and performed prediction. In order to test the accuracy of ANN prediction, internal validation will be made. Six experiments conducted and the mean square error used as a scale to measure the accuracy of prediction. First three experiments which with 50 input variables and one output used 80%, 60% and 40% of data for training. While the last three experiments change the number of input variables to 15 and use 80%, 60% and 40% of data for training. The results between the six experiments were compared. It was discovered that when the size of trained data reduced, the MSE value increased. In contrast, while the size of trained data increased, the MSE value decreased. Lower MSE value means better prediction. Overall, the accuracy of prediction for artificial neural network is high. The changes in the number of input variables will not affect the power of ANN prediction. However, quantity of data is one of the important factors that affect ANN prediction result. With larger data size, the ANN prediction is more accurate.