There is still a lack of automatic monitoring for freezing rain in the current meteorological observation field. The Random Forest (RF) machine learning method is introduced here, and an automatic recognition method for freezing rain is established using the data from Ka-band Millimetre-Wave Cloud Radar (MWCR). This provides a possibility to make up for the lack of automatic recognition and continuous observation of freezing rain. Firstly, the distribution characteristics of echo intensity and skewness values of Ka-band MWCR with different precipitation phenomena (rain, freezing rain, snow, and mixed rain and snow) during two freezing weather processes occurring in the Wuhan area in February 2024 are analysed. Significant differences in the value range and vertical height distribution are found. Then, echo intensity, skewness values, and near-surface temperature are determined as identification variables. Automatic recognition models for freezing rain using the RF machine learning method are established for several freezing processes in Wuhan and Guizhou respectively. After training and verification calculations, the test recognition accuracy rate (A_cc) exceeds 90%, and the freezing rain hit rate (P_od) exceeds 80%. Independent freezing rain examples are used for verification, and the verifying A_cc reaches 80%. Compared with wire icing observation, this method can automatically and continuously identify freezing rain phenomena in minutes, which is also feasible for business application. Since the echo intensity and skewness values of MWCR during the freezing rain process have obvious regional characteristics, freezing rain sample data from different regions should be collected to establish recognition models for different regions. The recognition accuracy can be improved by expanding samples and optimising model parameters and indicators, as well as reducing the False Alarm Rate (F_ar).