64th ISI World Statistics Congress - Ottawa, Canada

64th ISI World Statistics Congress - Ottawa, Canada

A Statistical Method for Predictive Analytics of Spatial Economic Development

Conference

64th ISI World Statistics Congress - Ottawa, Canada

Format: IPS Abstract

Session: IPS 152 - Statistics Concourse of Machine Learning and Artificial Intelligence

Monday 17 July 10 a.m. - noon (Canada/Eastern)

Abstract

Predictive analytics, as a stage of Gartner's model of “ascendancy analytics”, answers the question “Why did this happen?" based on an arsenal of supervised machine learning techniques since the analysis input contains labeled variables corresponding to the target indicators. The presence in the input data of variables that determine the geographical coordinates of the units of the population makes it possible to use spatial modeling methods in the analysis, first of all, Moran's spatial autocorrelation function. However, this raises the methodological problem of the randomness of the training and test samples. In most machine learning algorithms, these samples are formed as repeated random selection based on a table of random numbers. In this case, if units accidentally fall into these samples, the effect of their spatial correlation is violated. This technique was used and tested when applying the random forest method to solve the problem of multifactorial classification of cities in a number of countries in terms of economic and social development.