Cluster Analysis as Tool in Traffic Engineering

Regression analysis is a very common tool in traffic engineering analysis, partly because of the professional backgrounds of those doing the analysis but, perhaps, primarily because of an underlying premise that traffic can be described by deterministic models on which the observations have some randomness or imprecision superimposed. In this view, the “noise” can be filtered out by more observations and more precise observations, and the underlying deterministic model can then be identified. Another totally different underlying premise is that there are a finite number of distinct “states” or conditions in which the system may rest, with significant truly random deviations around each such equilibrium. If this premise is adopted, regression analysis is not a tool suitable for analysis and can provide misleading insights and meaningless relations. This paper applies the tool of cluster analysis to a set of traffic engineering data (specifically, left-turn factors in shared lanes) in which deterministic modeling and regression analysis have been applied in the past. Cluster analysis proved to be a powerful exploratory technique and helped identify several distinct modalities within the data. These can be explained by an underlying model based on a finite number of equilibrium states with significant random components or by an underlying model in which the process is truly stochastic and not at all deterministic, so that mild trends with considerable data scatter are to be routinely expected.