A DECISION SUPPORT SYSTEM FRAMEWORK FOR SEASONAL ZOONOSIS PREDICTION

The arising number of zoonosis epidemics and the potential threat to human
highlight the need to apply stringent system to contend zoonosis outbreak. Zoonosis is
any infectious disease that is able to be transmitted from other animals, both wild and
domestic, to humans. The increasing number of zoonotic diseases coupled with the
frequency of occurrences, especially lately, has made the need to study and develop a
framework to predict future number of zoonosis incidence. Unfortunately, study of
literatures showed most prediction models are case-specific and often based on a
single forecasting technique.
This research analyses and presents the application of a decision support system
(DSS) that applied multi forecasting methods to support and provide prediction on the
number of zoonosis human incidence. The focus of this research is to identify and to
design a DSS framework on zoonosis that is able to handle two seasonal time series
type, namely additive seasonal model and multiplicative seasonal model. The first
dataset describes the seasonal data pattern that exhibited the constant variation, while
the second dataset showed the upward/downward trend. Two case studies were
selected to evaluate the proposed framework: Salmonellosis and Tuberculosis for
additive time series and Tuberculosis for multiplicative time series. Data was
collected from the number of human Salmonellosis and Tuberculosis incidence in the
United States published by Centers for Disease Control and Prevention (CDC). These
data were selected based on availability and completeness.
The proposed framework consists of three components: database management
subsystem, model management subsystem, and dialog generation and management
subsystem. A set of 168 monthly data (1993–2006) of Salmonellosis and Tuberculosis
was used for developing the database management subsystem. Six forecasting
methods, including five statistical methods and one soft computing method, were
applied in the model management subsystem. They were regression analysis, moving average, decomposition, Holt-Winter’s, ARIMA, and neural network. The results of
each method were compared using ANOVA, while Duncan Multiple Range Test was
employed to identify the compatibility of each method to the time series. Coefficient
of Variation (CV) was used to determine the most appropriate method among them. In
the user interface subsystem, “What If” (sensitivity) analysis was chosen to construct
this component. This analysis provided the fluctuation of forecasting results which
was influenced by the changes in data. The sensitivity analysis was able to determine
method with the highest fluctuation based on data update. Observation of the result
showed that regression analysis was the fittest method for Salmonellosis and neural
network was the fittest method of Tuberculosis. Thus, it could be concluded that
results difference of both cases was affected by the available data series. Finally, the
design of Graphical User Interface (GUI) was presented to show the connectivity flow
between all DSS components.
The research resulted in the development of a DSS theoretical framework for a
zoonosis prediction system. The results are also expected to serve as a guide for
further research and development of DSS for other zoonosis, not only for seasonal
zoonosis but also for nonseasonal zoonosis.