Towards an adaptive multi-agent architecture for association rule mining in distributed databases

Association rule mining, which is a data mining technique, finds interesting association or correlation relationships among a large set of data items. Current association rule mining tasks can only be accomplished successfully only in a distributed setting, which will require integration of knowledge generated from the multiple data sites. Most existing architectures for mining in such circumstances require massive movement of data resulting in high communication overheads leading to slow response time. These challenges are heightened when we have extremely large data sizes in multiple heterogeneous sites. Moreover, most existing algorithms and architectures are only moderately suitable for real-world scenarios. There is therefore an urgent need for improved architectures that will explore the capabilities of software agents' paradigms in order to improve on the existing systems. This work therefore introduces an adaptive architectural framework that mines association rules across multiple data sites, and more importantly the architecture adapts to changes in the updated database giving special considerations to the incremental database with the X-Apriori algorithm. The results integration agent also adapts to changes in the results sites considering the data size; size of the agents; size of intermediate results; bandwidth, and other computational resources at the data servers. The proposed system promises to reduce communication and interpretation costs, improve autonomy and efficiency of distributed association rule mining tasks.