Spatio-temporal association rule mining framework for real-time sensor network applications

In this paper, we present a data mining framework to estimate missing or corrupted data in sensor network applications - a frequently occurring phenomenon in this domain. The framework is naturally germane to the spatio-temporal analysis of relational data stream evolution. Our method utilizes association rules to capture spatio-temporal correlations in multivariate, dynamically evolving, and unbounded sensor data streams. Existing approaches that tackled this problem do not account for the multi-dimensionality of the node data and their relationship; furthermore they entail simplistic and/or premature assumptions on the temporal and spatial factors to overcome the complexity of the streaming environment. Our technique, called Mining Autonomously Spatio-Temporal Environmental Rules (MASTER), comprehensively formulates the problem of mining patterns in sensor data streams, and yet remains provably adaptive to bounded time and space costs while probabilistically assuring a bounded estimation error. Simulation experiments show MASTER's efficiency in terms of overhead as well as the quality of estimation.