Positioning accuracy of global positioning system (GPS) is deteriorates when non-line-of-sight (NLOS) GPS multipath signals are received in urban environments. Therefore, it is important to determine GPS satellite visibility and reject NLOS satellites from positioning calculation to improve positioning accuracy. In this paper, we focus on a GPS correlation waveform which is affected by multipath signals. To detect NLOS multipath signals, we use machine learning of GPS correlation waveforms. We use neural networks for machine learning to generate the NLOS discriminator. From the evaluations of proposed method, the accuracy of GPS satellite visibility determination by using the neural networks is 96.8 %. The positioning accuracy without NLOS signals is also improved compared with the conventional positioning method.