Although RFID technology has been widely applied in the industrial fields, especially in logistics, it was restrictively implemented mainly because of detection problem, especially in environments of metals, farm products or liquids. We cannot overlook this problem because the RFID tag may be attached to the packages that contain materials which may influence the detection. Therefore, we need a way to predict the reading rate of RFID on various conditions and adequately cope with detectability problems. The purpose of this paper is to develop the prediction model for the RFID reading rate on various conditions. At first, we selected a list of test materials from expert interview and literature survey, and performed experimental design using orthogonal array table to investigate influence by materials. Second, we built prediction models using neural networks and support vector machine (SVM) with data from experiment. Finally, we compared the developed prediction models and selected the SVM model as the best prediction model. The SVM model gives us predicted reading rates of high accuracy under the conditions of various contents and environmental factors.