Discovering temporal rules from temporally ordered data

Abstract

We introduce a method for finding temporal and atemporal relations in nominal, causal data. This method searches for relations among variables that characterize the behavior of a single system. Data are gathered from variables of the system, and used to discover relations among the variables. In general, such rules could be causal or acausal. We formally characterize the problem and introduce RFCT, a hybrid tool based on the C4.5 classification software. By performing appropriate preprocessing and postprocessing, RFCT extends C4.5’s domain of applicability to the unsupervised discovery of temporal relations among temporally ordered nominal data.