FEATURE SELECTION METHOD BASED ON GENETIC ALGORITHM WITH WRAPPER-EMBEDDED TECHNIQUE FOR MEDICAL RECORD CLASSIFICATION

Abstract

Medical records, also known as Electronic Health Records, can be used to analyze medical data which contributes to greater knowledge of information related to disease, disease phase, and early identification of disease. The main problem in medical records is how to determine the features of the selected subset to make predictions on clinical data. The problem of classifying medical record data involves many features. This creates problems in determining which features have a correlation to the predicted results. Feature selection is an appropriate technique for selecting risk factors for disease in medical records. Feature selection requires optimization techniques so that the accuracy value can be increased. Therefore, feature selection using genetic algorithms has the ability to optimize the selected features. The feature selection technique, namely the wrapper, applies a learning algorithm to test the features to be selected. The embedded engineering approach to feature selection makes it possible to apply learning in the selection process. Genetic algorithm-based feature selection method with wrapper-embedded technique is expected to produce an effective feature subset. This technique evaluates features by using combinations in fitness calculations with the aim of assessing the selected features. The development of genetic algorithms by applying fitness value evaluation and elitist aims to maintain the fitness value in the next generation.