Kinerja Pendekatan Convolutional Neural Network dan Dense Network dalam Klasifikasi Citra Malaria

Abstract

Indonesia is an archipelago, which three of its five main island consists mainly, or dense tropical rainforest. This rainforest is main breeding ground for malaria disease that mostly affect regions near said forest. In an effort to treat malaria disease, a diagnostic process is performed to correctly identify the disease. Several image pattern recognition technique been developed and have potential to be utilized as malaria diagnostic tool. In this research, a method is described on designing neural network to detect a blood cell parasitized by malaria. The method consists of utilizing a dense network, and a convolutional neural network, to be trained using publicly available training dataset. Both models’ performance is then compared and analyzed. Before the data is used, a process of padding is performed to resize the input image into 200 x 200 pixels. The resized input data is then used to train both models. From the training and testing, it is found that the dense network achiever 64.78% accuracy. On the other hand, model based on convolutional neural network achiever 94.32%. From analysis, it is found that the size of the model being used is not big enough to achieve better performance. Hence, it is suggested for future research to increase the model size in terms of network width and depth.