In order to improve the flame retardancy of polypropylene/magnesium hydroxide (PP/MH) composites and to grasp the intensity of different influencing factors on the flame retardancy of the materials, the MH particle size, contact angle, and the amount of additive were used as the three inputs, and the limiting oxygen index (LOI) of PP/MH composites was taken as the output. A three-layer BP neural network prediction model was established, and the orthogonal test results were used as samples to train it to predict the flame retardancy of the composites. The prediction results were verified by experiments. The results show that the effects of various factors on the flame retardancy of the PP/MH composites from large to small are MH content, MH contact angle and MH particle size. The optimal process parameters: the MH particle size is 0.2 µm, the MH contact angle is 135°, the MH content is 40%. Under these conditions, the LOI of PP/MH composite is as high as 31.5%. The BP neural network model can accurately predict the flame retardancy of composites, and the relative error between the predicted value and the experimental value is generally less than 5%. The prediction model of flame retardancy can be used to optimize the performance of materials, reduce the experimental workload, and improve the work efficiency.