ABSTRACT
Forecasting public revenues at the municipal level is
quite relevant for public administration in Brazil since
it defines the budget of public policies. However, it is a
complex process, and its consequences that permeate
the divergence between executed and budgeted values
have harmful effects on the population. So, aiming to
mitigate this problem, we analyzed to optimize public
revenue forecasting using feature selection through
model combination (ensemble) of wrapper filter
approaches. We assessed the proposed solution in light
of the Brazilian Fiscal Responsibility Law, using São
Luís, Maranhão, as a case study. Our results suggested
that our approach is suitable for reducing complexity
and improving forecasting performance. The method
can be used in other Brazilian state capitals, boosting
data-driven public policy-making.