Classification of pneumonia patients risk using hybrid genetic algorithm-discriminant analysis and NaÏve Bayes

Pneumonia is the most common causes of death in developing countries, such as in Indonesia. Therefore, appropriate pneumonia classification is very important in determining the disease severity and to know the most appropriate treatment for the patient. In this study, Discriminant Analysis (DA), hybrid Genetic Algorithm- Discriminant Analysis (HGA-DA) and Naïve Bayes (NB) are used to classify risk class of patient. GA is an artificial intelligent method that can avoid a trap in local optima and easy to implement in solving various objective functions and constraints, while NB is a simple but powerful method that returns not only prediction but also the degree of certainty. In this study, GA is used to improve multi-class classification performance of DA. Firstly, GA is used for variable selection in DA, and then a comparative study with other variable selection methods is performed. In addition, Genetic Algorithm is also implemented for parameter estimation. Analysis results show that there are differences in selected variables from four selection methods in classifying patient risk class. The use of hybrid methods of DA and GA in variable selection and parameter optimization stages gives better multi-class classification results than DA or NB, since it produces highest Geometric Mean (GM) and Area Under Curve (AUC) criterion.