Association rule mining with mostly associated sequential patterns

In this paper, we address the problem of mining structured data to find potentially useful patterns by association rule mining. Different than the traditional find-all-then-prune approach, a heuristic method is proposed to extract mostly associated patterns (MASPs). This approach utilizes a maximally-association constraint to generate patterns without searching the entire lattice of item combinations. This approach does not require a pruning process. The proposed approach requires less computational resources in terms of time and memory requirements while generating a long sequence of patterns that have the highest co-occurrence. Furthermore, k-item patterns can be obtained thanks to the sub-lattice property of the MASPs. In addition, the algorithm produces a tree of the detected patterns; this tree can assist decision makers for visual analysis of data. The outcome of the algorithm implemented is illustrated using traffic accident data. The proposed approach has a potential to be utilized in big data analytics.