FORECASTING GDP GROWTH USING DATA MINING ON THE EXAMPLE OF SERBIA

Abstract

Predicting the outcome of various phenomena has always been an attractive research topic in a large number of scientific disciplines, especially in economics. As a scientific discipline, econometrics provides various models for predicting indicators such as GDP, inflation rate, interest rate, price of various goods and services, as well as many others at both micro and macro levels. The development of information technologies has made computational operations much faster and more precisely. However, a special contribution is reflected in the application of data mining for the purpose of extracting relevant information from a large data set. Models developed using data mining provide good results in predicting economic indicators, often more successfully than certain econometric models. This paper aims to forecast the growth of GDP through the application of time series mining on the example of the Republic of Serbia. The analysis was performed in two cases: in the first one models include independent attributes that additionally describe the dependent variable, while in the other case they do not contain these attributes. Three different mining methods were used in both cases (linear regression, multilayer perceptron and random forest) and the obtained results of model validation were presented and interpreted.