A Decision Support System for Municipal Budget Plan Decisions - New Contributions in Information Systems and Technologies - page 129

choosing the best model is the measurement error closer to zero. Error measures used
were the root mean square error and the average percentage error.
From the applied models the one which had, in most indicators, the closer to zero
error was classical decomposed with linear trend (tslm() function) as can be seen from
the results presented in Table 3. Is important to notice that the major objective of the
forecast model is not to obtain accurate predictions, but more to learn the
characteristics of the several indicators and their evolution in the last few years. This
is fully accomplished with this conclusion of that the majority had a linear trend and
none has any observable seasonality factors. This is also the reason why was not
considered the use of techniques that combine predictions like meta-learning
algorithms as they lose the descriptive aspects of the model [12].
We also used the degree 3 polynomial model, the splinef() function [10], but they
proved inadequate because of big errors in comparison with other models showing,
therefore, a poorer fit. The likely reason for the poor performance of these models is
due to the fact that the estimates only take into account the last two years, and all
other points, however well-adjusted are ignored. Thus, these models may be useful for
modeling series with seasonality, using classical decomposition methods, but are of
little use to model series with only trend.
Some parts of the system, such as the database, the algorithms of data collection
and storage, are already implemented and fully functional. This was paramount for
getting data, allowing the study on prediction models. Consequently, with this DSS,
decisions are taken considering the enrichment of indicators, which is an asset to the
management of the municipality.
It is important to recognize that such a DSS system is never finished, requiring
constant monitoring and updating.
From a future perspective, user levels should be set, as well as the development of
a proper version of the application to mobile devices. Similarly, over time, the
collection of new values making the time series longer, could allow the estimation
and use of more sophisticated models, for example, those based on neural networks as
used by Cortes [13].
Acknowledgments. This article would not have been possible without the help of
those who provided us with the means, the data and the support. Thus, we emphasize
the participation of the Lagoa Municipality Mayor, Eng. João Ponte, the Head of
General Administration Division, Drª. Clara Ganhão and other colleagues of the first
author, for their patience and availability shown in the prompt clarification of all
issues.
References
1. Valacich, J., Schneider, C.: Information Systems Today. Prentice Hall (2013)
2. Burstein, F., Holsapple, C.: Handbook on Decision Support Systems. Springer (2008)
3. Power, D.J.: Decision Support Systems: Concepts and Resources for Managers.
Greenwood/Quorum, Westport (2002)