Bagian 2: Model Arsitektur Neural Network Dengan Kombinasi K-Medoids Dan Backpropagation Pada Kasus Pandemi COVID-19 Di Indonesia

Abstract

The aim of the research is to create a prediction model on the best neural network architecture by combining the k-medoids and backpropagation methods in the case of the COVID-19 pandemic in Indonesia. Data obtained from the Ministry of Health is sampled and processed from covid19.go.id and bnpb.go.id. The case raised was the number of the spread of the COVID-19 pandemic in Indonesia as of July 7, 2020, with 34 records. The variables used in this study are the number of positive cases (x1), the number of cases cured (x2), and the number of deaths (x3) by province. The process of data analysis uses the help of RapidMiner software. The solution provided is to combine the k-medoids and backpropagation methods. Where the k-medoids method is mapping the specified cluster. The cluster labels used are high cluster (C1 = red zone), alert cluster (C2 = yellow zone), low cluster (C3 = green zone). The results of cluster mapping are continued to the backpropagation method to predict the accuracy of the existing cluster results. By using the best architectural model 3-2-1, the accuracy value is 94.17% with learning_rate = 0.696. Cluster mapping results obtained nine provinces are in the high cluster (C1 = red zone), three provinces are in the alert cluster (C2 = yellow zone), and 22 provinces are in the low cluster (C3 = green zone). It is expected that the results of the research can provide information to the government in the form of cluster mapping of regions in Indonesia.