A Comparative Study on Different Feature Selection Methods for Malaria Detection
Fatma Günseli Yaşar Çıklaçandır, Özlem Karabiber Cura · 2023
Early diagnosis and treatment of malaria disease, which has a risk of death, is important. The importance of time has led to the emergence of automation studies in this field. The features drawn from the layers of pool5 and fc1000 of ResNet18 are selected using various feature selection techniques (chi square, MRMR (minimum redundancy maximum relevance), ReliefF, Flap, F test). During the evaluation of the best feature selection approach, it is discovered that the features drawn from pool5 layer give successful results when feature selection is applied with MRMR. When the features captured in different numbers using MRMR are classified, successful results are obtained even when the number of features is 100 or 250. Thus, it has been seen that the features drawn from the pool5 layer of ResNet18 can be classified in a shorter time by reducing them to a smaller number of features thanks to MRMR.