In this decade, the internet becomes indispensable in companies and people life. Therefore, a huge quantity of data, which can be a source of hidden information such as association rules that help in decision-making, is stored. Association rule mining (ARM) becomes an attractive data mining task to mine hidden correlations between items in sizeable databases. However, this task is a combinatorial hard problem and, in many cases, the classical algorithms generate extremely large number of rules, that are useless and hard to be validated by the final user. In this paper, we proposed a binary version of grey wolf optimizer that is based on sigmoid function and mutation technique to deal with ARM issue, called BGWOARM. It aims to generate a minimal number of useful and reduced number of rules. It is noted from the several carried out experimentations on well-known benchmarks in the field of ARM, that results are promising, and the proposed approach outperforms other nature-inspired algorithms in terms of quality, number of rules, and runtime consumption.