Machine Learning and Statistical Models for the Prevalence of Multiple Sclerosis

摘要

Multiple sclerosis is an immune-mediated disease affecting approximately 2.5 million people worldwide. Its cause is unknown and there is currently no cure. MS tends to be more prevalent in countries that are farther from the equator. Moreover, smoking and obesity are believed to increase the risk of developing the disease. This article builds machine learning and statistical models for the MS prevalence in a country in terms of its distance from the equator and the smoking and adult obesity prevalence in that country. To build the models, the center of population of a country is approximated by finding a point on the surface of the Earth that minimizes a weighted sum of squared distances from the major cities of the country. This study compares the predictive performance of several machine learning models, including first and second order multiple regression, random forest, neural network and support vector regression.