Referring to the doc. link from you.
WRED
Cisco's implementation of Random Early Detection (RED), called Weighted Random Early Detection (WRED), differs from other congestion-avoidance techniques because it attempts to anticipate and avoid congestion, rather than controlling congestion when it occurs.
WRED takes advantage of the Transmission Control Protocol (TCP) congestion control to try to control the average queue size by indicating to end hosts when they should temporarily stop sending packets. By randomly dropping packets before periods of high congestion, it tells the packet source to decrease its transmission rate. Assuming the packet source is using TCP, WRED tells it to decrease its transmission rate until all the packets reach their destination, meaning that the congestion is cleared.
WRED reduces the chances of tail drop by selectively dropping packets when the output interface begins to show signs of congestion. By dropping some packets early rather than waiting until the queue is full, WRED avoids dropping large numbers of packets at once. Thus, WRED allows the transmission line to be fully used at all times. WRED also drops more packets from large users than small. Therefore, sources that generate the most traffic are more likely to be slowed down versus sources that generate little traffic.
You can enable WRED and configure the two threshold percentages assigned to the four egress queues on a Gigabit-capable Ethernet port by using the wrr-queue random-detect max-threshold interface configuration command. Each threshold percentage represents where WRED starts to randomly drop packets. After a threshold is exceeded, WRED randomly begins to drop packets assigned to this threshold. As the queue limit is approached, WRED continues to drop more and more packets. When the queue limit is reached, WRED drops all packets assigned to the threshold. By default, WRED is disabled.
You modify the DSCP-to-threshold map to determine which DSCPs are mapped to which threshold ID by using the wrr-queue dscp-map interface configuration command. By default, all DSCPs are mapped to threshold 1, and when this threshold is exceeded, all the packets are randomly dropped.
If you use WRED thresholds, you cannot use tail drop, and vice versa. If WRED is disabled, tail drop is automatically enabled with the previous configuration (or the default if it was not previously configured).
Queueing
You assign two drop thresholds to each queue, map DSCPs to the thresholds through the DSCP-to-threshold map, and enable either tail drop or WRED on the interface. The queue size, drop thresholds, tail-drop or WRED algorithm, and the DSCP-to-threshold map work together to determine when and which packets are dropped when the thresholds are exceeded. You configure the drop percentage thresholds by using either the wrr-queue threshold interface configuration command for tail drop or the wrr-queue random-detect max-threshold interface configuration command for WRED; in either case, you map DSCP values to the thresholds (DSCP-to-threshold map) by using the wrr-queue dscp-map interface configuration command.
It does drop the packets randomly if reach the threshold and congestion. What area you do not undetstand ?
