Association rule mining is one of the most important issues in data mining. Apriori computation schemes greatly reduce the computation time by pruning the candidate itemset. However, a large computation time is required when the treated data are dense and the amount of data is large. With apriori methods, the problem of becoming incomputable cannot be avoided when the total number of items is large. On the other hand, bottom-up approaches such as artificial life approaches are the opposite to of the top-down approaches of searches covering all transactions, and may provide new methods of breaking away from the completeness of searches in conventional algorithms. Here, an artificial life data mining technique is proposed in which one transaction is considered as one individual, and association rules are accumulated by the interaction of randomly selected individuals. The proposed algorithm is compaired to other methods in application to a large-scale actual dataset, and it is verified that its performance is greatly superior to that of the method using transaction data virtually divided and that of apriori method by sampling approach, thus demonstrating its usefulness.