Anemia Detection Through Conjunctiva on Eyes Using Histogram of Oriented Gradients and Decision Tree Classification

Generally, an invasive examination is carried out by taking a sample of a person's blood to detect hemoglobin levels. However, as science develops, hemoglobin levels can be measured through alternative non-invasive examinations by looking at the paleness of the eye's conjunctiva. This research aims to detect anemia through images of the eye's conjunctiva with the histogram of oriented gradient (HOG) characteristics extraction method and decision tree also logistic regression comparison classification methods. Based on the test results using HOG and decision trees on 40 testing images, with an image size of 256x128, the best parameters were obtained, namely cell size of 8x8, block size of 8x8, bin numbers of 15, normalization of L2-Hys blocks, entropy criteria, best splitter, random state 10, and minimum impurity decrease of 0.15, the system was able to produce the largest performance of 82.5% with a computing time of 25.63 seconds. Meanwhile, for logistic regression method with a stochastic gradient descent (SGD) optimizer methods, the best parameters are cell size of 16x16, block size of 8x8, bin numbers of 11, block normalization of L2-Hys, and random state 30, the system is able to produce the largest performance of 92.5% with a computational time of 24.20 seconds.