Abstract:

Nowadays, it has become very difficult for clinical decision support systems to obtain useful information from very large medical databases. Genetic algorithms (GA) are a widely used feature selection method and can give the best solutions. In this study, a model with GA is proposed to select features from medical data with a large number of complex data and to increase classification success by creating the most appropriate feature subset. In order to evaluate the performance of the proposed method, 5 most well-known and easily accessible medical data sets and 7 different supervised classification methods were used in the study. Feature selection and classification applications were made separately for each data set and each classifier. The results obtained in these applications revealed that, depending on the data set, in the classifications made with the proposed approach, an average of 2% to 21% increase was achieved in the accuracy rate and thus in the machine learning model performance. In addition, it is seen that the Random Forest, one of the supervised classification algorithms, gives better results in all data sets than other algorithms and it has been seen that it stands out with its classification success in medical datasets.

Keywords: