Sensor time series association rule discovery based on modified discretization method

Association rule discovery from sensor time series is a challenge. Because the time series has high dimensional, numerical and continuous nature. However the general association methods can only deal with data which are symbolic and discrete. And the general association methods have high processing time consumption when the data have high dimension. So a useful framework is proposed, which is pre-processing, representation, discretization and temporal association mining. In the discretization section, a modified discretization method is proposed which can combine the advantages of other methods, such as piecewise aggregate approximation (PAA), knee point selection, symbolic aggregate approximation (SAX) and monotonicity feature extraction. In the association section, a modified Apriori algorithm is proposed to discover special patterns and normal rules.