The tactile internet that will enable human tactile to visual feedback control has drawn significant attentions from both academic and industrial communities. One of the key challenges of tactile internet is how to ensure ultra- reliable and low-latency communications between the local operator and the remote device. Existing studies found that the packet arrival process in tactile internet is very bursty. However, how to design burstiness aware bandwidth reservation for tactile internet is still unclear. To this end, we first apply Neyman-Pearson method to classify the arrival process of each user in high and low traffic states, and then optimize reserved bandwidth for them. The optimal overall bandwidth required to guarantee the latency and reliability requirements is achieved. Classification errors, queueing delay violation, and decoding errors are considered for high traffic users. Transmission collisions in random access procedure and decoding errors during retransmission phase are taken into account when users are in low traffic state. Simulation results show that the proposed method can save up to $66.7\%$ of the bandwidth compared with the method that is not aware of burstiness.