Abstract:
Abstract
Introduction: - Malaria is continues to be a leading cause of morbidity and mortality
worldwide. In Ethiopia, transmission is unstable and seasonal, with occasional devastating
epidemics. In health facilities morbidity data can be used for prediction of occurrence of
disease and can help in decision making using data mining techniques.
Objective: - The main objective of this study was to predict malaria species morbidity from
malaria data by using data mining technique in Chewaka Health Center, South-West
Ethiopia, 2012.
Methods: Institution based retrospective record review study was conducted. All malaria
positive data in Chewaka health center from September 01, 2007 to December 30, 2011
was collected from manual records in to new format prepared for the study purpose and
the data was integrated with meteorology data of nearby meteorology station. Data quality
was assured by using cross-check up of data collected with manual available record. A
total of 5077 records were used and data analysis was done by using WEKA classification
decision tree J48 and neural network algorisms with two modes, 10 fold cross-validation
and 90%-10% percentage split for training-testing modes.
Results: Of 5077 records in dataset class attribute accounts P. faliciparum 2745(54.1%),
P. vivax 2258(44.5%) and mixed 74 (1.5%). Prediction model developed by Decision tree
J48 algorism with 90%-10% training-test mode was scored the highest accuracy and
selected as best model. The model predicted correctly 86.22 % P. faliciparum, P. vivax
86.14 %and as mixed 99.4% species in their class, over all predictive accuracy was
90.5%. Mean monthly relative humidity, mean monthly maximum temperature, total
monthly rainfall, Age and address were selected by the models as best predictive
attributes for malaria species.
Conclusion and recommendation:-This study showed malaria morbidity to be predicted
by mean monthly relative humidity, mean monthly maximum temperature, total monthly
rainfall, Age and address. Using meteorology and morbidity data mining that can help for
effective decision making on malaria prevention and early warning of epidemics is
recommended.
