Forecasting Dengue Hemorrhagic Fever Incidents: A Machine Learning Approach
Tien Rahayu Tulili(*a), Yohanes K. Windi (b), Bambang Cahyono(a), Damar Nurcahyono(a), Karyo Budi Utomo(a), Ahmad Rofiq Hakim(a)

(*a) Department of Computer Engineering, Politeknik Negeri Samarinda,
Jl. Dr. Ciptmangunkusumo Kampus Gunung Panjang, East Kalimantan, Indonesia.
*tien.tulili[at]polnes.ac.id
(b) Politeknik Kesehatan Kemenkes
Jl. Pucang Jajar Tengah 56, Surabaya, Jawa Timur, Indonesia

Abstract

Dengue is a viral infection transmitted by Aedes mosquitos. This disease mostly spread in the tropical and sub-tropical countries and according to WHO, the dengue outbreaks has increased 30-fold over the last five decades. The disease is still an ongoing burden of throughout the world. In Indonesia, for example, the incident of dengue hemorrhagic fever (DHF) has shown up 8,056 cases spread in the last five years. One of the ways to help the government to mitigate any possible of the spread is by utilizing a nearly accurate forecast system in predicting the cases. This study aims to develop machine learning as the most accurate predicting method of DHF cases in East Kalimantan. Various kinds of data are used in developing some machine learning models. Furthermore, identifying variables prior the models^ development is done to achieve the best model of prediction- furthermore, a comparative study of the models built is discussed.
Monthly dengue cases, incidence rate (IR), climate factors (rainfall, atmospheric pressure, the duration of the sun) and socio-economic conditions (population density, the number of inhabitants) from three different cities/districts (Samarinda, Balikpapan, and Berau) in East Kalimantan from 2007-2019 become data of the study. Those data are collected from the Local Department of Ministry of Heath Republic Indonesia, and Indonesia Central Bureau of Statistics (CBS). Prior machine learning^s modeling, all data are analyzed with Pearson Correlation method to identify which variables has a positive correlation with DHF cases. Several machine learning algorithms, those are: Neural Network, Deep Learning, Generalized Linier Model, Generated Boast Tree and KNN, implemented in the modelling and forecasting. The results showed that some climatic factors are negatively correlated to DHF cases in East Kalimatan. Furthermore, the best method for forecasting the incidence is neural network with the RMSE value was 8.660%.

Keywords: forecast, dengue hemorrhagic fever, machine learning, deep learning, neural network, generalized linier model, and KNN

Topic: Engineering