A Decision Support System for Municipal Budget Plan Decisions - New Contributions in Information Systems and Technologies - page 128

Table 3. Error measures for the forecasting models of ten indicators (own construction)

ets() tslm() auto.arima()
RMSE APE RMSE APE RMSE APE
Capital Expenses 2 480 746 -151,7 2 126 959 -16,18 1 823 987 -27,57
PPI Execution 13,61 -9,37 355 911,7 -120,81 13,61 -9,37
Direct Taxes 378 887,5 -4,04 287 742,5 -4,79 348 677,7 -1,41
Indirect Taxes 378 887,5 -4,04 61 297,95 -63,18 62 308,07 -62,72
Government (in)Direct Taxes 411 744,9 -7,10 344 925,4 -5,21 373 303,3 -0.94
Current Revenue 542 753,6 -3,15 493 621,5 -1,09 509 867 0,26
Municipal Taxes, Penalties,... 40 519,34 -14.32 46 391,1 -12.38 38 075,88 -6.72
Current Transfer 270 004,3 -192,26 259 061,1 -157,17 269 967,9 -190.34
Capital Transfers 8 413 426 -111,87 969 969 -54,75 1 123 014 -91,69
Capital Income 1 768 804 -37,50 1 666 251 -21,20 1 672 661 -21.52
In short, according to the calculated error values, it seams that the model with
better results is the classical decomposition model built by the tslm() function, one of
the simplest models. However, the other two methods cannot be discarded since they
may also constitute a good forecast method for some indicators.

4 Conclusions

In an organization, management decisions are based on guidelines, provided by the
specialized areas. In this work, a study for the development of a Decision Support
System to provide indicators, which support the implementation of management
decisions in the context of a local government organization, the municipality of Lagoa
(São Miguel, Azores) is presented.
A literature review was made on the core issues necessary for the development of
this project, including decision support systems, forecasting models, as well as some
practical cases. Then, a characterization of the current system, procedures and
workflow, through UML diagrams were made.
The new system is featured with UML modeling and implemented as a MySQL
database management system, and also, data collection algorithms with scripts in
PHP, an application of predictive models in R, and finally the results discussion.
For the prediction, only twelve annual data points were available. The lack of data
would derive the project to a model base DSS as an expert system. But, son we realize
that there were no relevant knowledge on the organization about the behavior of these
indicators, and so, we could only obtain same information from the small data set. Is
also relevant to verify the minimum sample size to obtain accurate forecasts is highly
dependent on data noise, which is not a problem with the data used [11].
In this way, several models were tested, such as, exponential smoothing, classical
decomposition with linear trend, and ARIMA models. The decision criterion for