Cellular communication systems aim at achieving complex largescale monitoring and tracing applications in wider fields, which is based on mobile station nodes localization. By studying the existing node localization technologies, this paper analyses the current several classical machine learning algorithms purposefully, and proposes a cellular communication system node localization algorithm based on machine learning, using it as a centralized coordinate algorithm of distributed node localization. Through simulation and theoretical  analysis, it proves that the node localization algorithm in cellular communication systems based on machine learning can resolve the border problem and the coverage hole problem in the traditional algorithms based on signal parameters, and its overall function is better than the traditional algorithms based on signal parameters and GPSone in terms of average error, standard deviation and the accuracy rate of distributed localization as well as the cost superior to the traditional location algorithm based on signal parameters.