Apriori algorithm can generate huge sets of decision rules. A lot of these rules, especially in a multilevel multidimensional environment, are redundant or useless. To reduce this unwanted effect the author proposes to utilise generators or closed itemsets, originally used in lossless itemsets representations, to filter decision rules during their generation. These two modifications of the Apriori algorithm enable reduction of a size of resultant decision rules' sets without lost of information about relations in training data. Choice of the proposed modifications depends on user's preference about length of the rules. Filtering using generators gives shorter rules while filtering with closed itemsets results in smallest sets of rules. Both modifications were evaluated on large sets of training data from faults simulations experiments.