Using soft set theory for mining maximal association rules based on the concept of frequent maximal itemsets which appear maximally in many records has been developed in recent years. This method has been shown to be very effective for mining interesting association rules which are not obtained by using methods for regular association rule mining. There have been several algorithms developed to solve the problem, but overall, they retain weaknesses related to the use of memory as well as mining time. In this paper, we propose an effective strategy for maximal rules mining based on soft set theory that consists of the following steps: 1) Build tree Max_IT_Tree where each node contains maximal itemsets X, the category of X, the set of transactions in which X is maximal, and the support of the maximal itemsets X for each category. 2) From the tree Max_IT_Tree built in previous steps, build a tree Max_Item_IT_Tree so that each maximal itemset has child nodes where each node contains items with categories different from the category of maximal itemsets. 3) Generate maximal association rules which satisfy predefined minimum M-support (min M-sup) and minimum M-confidence (min M-conf) thresholds.