A Comparative Study of Multiclass Classification Method for Mapping COVID-19 Risk Zone in Java Island

Various attempts are needed to control the increment of COVID-19 cases in Indonesia, especially Java Island. One of the effective attempt to do this is through the preventive act by providing news about a region. Indonesia, through Satgas Penanganan COVID-19, has built a risk zone of district/city as a warning system for the public and the substance of policy making for government in region level. The risk zone is built by three kinds of indicator using a conventional technique named score weighting. By considering the importance of the risk zone for policy making in the government, this study aims to build a risk zone classification model for districts / cities in Java using several data mining classification techniques and determine the best classification model based on evaluation results. This study uses several classification technique on the purpose of comparation. These techniques are naive Bayes, decision tree, k-nearest-neighbor, and neural network. Before entering the modeling stage, data is being adjustedat the preprocessing stage where missing value and imbalanced data problems are identifies. These problems is being overcome by doing data imputation and oversampling techniques. The result of this study indicates that k-nearest-neighbor is the best model compared to other three models. This result is based on the evaluation measures of the four models where the k-NN model has the highest accuracy value, the macro average value for sensitivity, specivicity, and F1-Measure compared to other models