Predictive Models of Economic Systems Based on Data Mining

Data election to build a representative model able to explain socio-economic phenomena is a challenge within the model construction stage itself. Knowing what data to include within the studies and what to discard is a challenge, and again, at the same time, a great amount of possible factors affecting each variable behavior must be found. In complex phenomena, the number of factors affecting a variable is enormous, and isolating a variable can become a hopeless effort. Besides, there are also factors that are difficultly observable or inherently not observable that must be considered, those ones known as errors or perturbations in a relation that have influence in the constructed model outputs. Techniques applied in data mining can give support to the studies in the moment of analyzing the socio-economic phenomena and demonstrate results obtained through a scientific and reliable way. Data mining is proposed as a valid option in the study of indicators contrasting the traditional methodology (econometrics). An experiment was conducted to contrast two cultures in the use of statistical modeling. One assumes that the data are generated by stochastic GIVEN data model (Data Modeling Culture). The other one uses algorithmic models and treats the data as unknown mechanism (Algorithmic Modeling Culture).