For the applications of wireless sensor networks in typical industrial environments, the wireless link quality of IEEE 802.15.4 has a certain degree of fluctuation and packets loss appear bursty. In this paper, a data-trace-driven modeling method and statistical inference is adopted for the statistical analysis of the loss interval in link-layer packet transmission traces. The study finds that the loss interval follows Pareto distribution and the distribution parameter reflects the distribution characteristics of loss interval under a different packet loss rate. Take the D_α, namely correlation distance, which is the difference between the sample distribution parameter and that of the independent identical distribution under the same packet loss rate, as the indicator to characterize burstiness of packet loss. Through analysis and experiments, it is indicated that the retransmission independence distance (RID) in the link-layer has an approximate linear relation to the correlation distance. The study designs an optimal data retransmission mechanism using the indicator D_α in single-hop links, which can improve data retransmission reliability. Finally, it is verified by experiments that the average of reliability improvement is 6.3% with the maximum 14.5%, which is relative to the fixed time-slot retransmission mechanism that does not consider the packets loss burstiness.