If my memory serves, in a closed systems, there are many microstates (microscopic degrees of freedom that the time average of the state of the system can be with equal probability each (i.e. assuming ergodicity)) in the system.
By statistical arguments, if there is a larger number of microstates can contribute to the same macrostate (observed quantities of the system such as temperature, volume, energy, alignment of spins etc.), then over time, the system is very likely to be in states with maximum number of microstates, and hence entropy (ln of the number of microstates) is maximised