An Extension to the TIMERS Method

In this paper we present TIMERS II (Temporal Investigation Method for Enregistered Record Sequences II). Assuming that the effects take time to manifest, TIMERS II merges the input records and brings the causes and effects together. The output is in the form of a set of decision rules. The condition attributes’ values could have been observed in the past or the future relative to the decision attribute’s value. In TIMERS II the past values can influence the present, thus establishing causality. But we consider it possible to reference observations that appear after the decision attribute, which forms the basis for acausality, or temporal co-occurrence. Three tests are performed using three different assumptions on the nature of the relationship. Each test results in a number of classification rules. The quality of the output rulesets determine if the decision attribute's value is best described by a nontemporal (instantaneous) relationship, or a temporal (causal or acausal) one. In previous work, TIMERS would allow referencing the values of condition attributes that were all observed either after or before the decision attribute, and so the rules followed a unique direction in time. In this paper, we consider it possible to reference both the past and the future values in the same rule to predict an attribute's value. We also present a new algorithm that uses the accuracy of the generated rulesets to determine whether the relationship between the decision attribute and the condition attributes is instantaneous, causal, or acasual.