Feature selection based on JAYA algorithm for medical datasets

In recent years, due to the developments in technology, large amounts of data are emerging in every field. While the majority of this data is meaningless, only a small part of it is interrelated and meaningful. It can be very costly to produce consistent results from this huge amount of data. One of these fields is the medical field. Researchers have put forward many studies to produce consistent results in the medical field. In this paper, we propose a 2 step hybrid feature selection method that combines filter and wrapper feature selection methods for different medical datasets with a large number of attributes (except for one dataset, the others are larger than 2000). In the first step, Relieff's algorithm is used to filter out redundant attributes from the medical datasets and send them to the second step, the wrapper step. In this step, Knn (k-nearest neighbor) method was preferred as a classifier. In this algorithm, which we call JASAL, the Levy Flight method is added to strengthen the search capabilities of JASAL. The feature selection performance of the JASAL algorithm was tested with 8 different medical datasets. The results of the JASAL algorithm are compared with the feature selection performance of 11 different metaheuristic algorithms with the metrics of average fitness, standard deviation, accuracy, average feature selection size. According to the results, the JASAL algorithm achieved the highest accuracy. Moreover, it achieved this by selecting fewer features. Also, the average ranking values are better than the other algorithms. All the results show that the proposed method provides high classification accuracy on datasets with a very large number of attributes and has a significant ability to generate a smaller subset of attributes that are meaningful to each other.