Learning the Relationship between Asthma and Meteorological Events by Using Machine Learning Methods

Abstract

In this article, a new methodology is proposed by using the relationships between meteorological events and asthma cases of asthma patients in a region compared to other regions in a country. We focus on the impact of weather conditions on asthma in order to estimate asthma cases using machine learning methods based on meteorological events only. In order to increase the success of the estimates, in addition to the 10 features identified by the National Environmental Information Centers, we create some new semi-synthetic features by using the multiplication and addition operations on the features given after the scaling. Then, we use machine learning methods and the R-square coefficient approach to learn the effective features using the features obtained from publicly available data sets for Russia. After determining the effective features, we use three different machine learning algorithms: random forest, linear regression, and kernel ridge regression algorithms. We use transfer learning to store effective features obtained from a dataset for Russia and then apply them to a dataset for Kazakhstan. Our hypothesis is that a combination of the selected semi-synthetic properties of the random forest algorithm has the best performance accuracy for this application. The model successfully identifies (predicts) very high, high, medium, low or very low numbers of people with asthma for the first time in the region.