In the indoor visible light communications (VLC) network with multiple antennas and multiple users, in order to improve the time complexity of the optimal antenna selection algorithm under the condition of increasing number of transmitting antennas and users, the naive Bayes (NB) method is applied to the downlink light emitting diodes (LED) selection problem of the indoor multi-user VLC network. Firstly, the LED selection task is modelled as a multi-classification problem, and the training sample set is generated using the users' known channel state information. Then, the corresponding class labels are generated by maximizing the sum-rate of the multi-user VLC network. Futher, based on the generated training sample set, the classifier model is obtained using the NB method. Finally, the trained classifier model is applied to the LED selection of new users. Simulation results show that, compared with the optimal LED selection algorithm of the multi-user VLC network, the proposed LED selection scheme based on the NB method can effectively reduce the time complexity and achieve a good balance between algorithm complexity and system sum-rate.