USE OF FUZZY INFERENCE FOR MODELING PREDICTION OF TRANSIT RIDERSHIP AT INDIVIDUAL STOPS

Document Type

Journal Article

Publication Date

2001

Subject Area

infrastructure - stop, ridership - commuting, ridership - forecasting, ridership - forecasting, ridership - demand, mode - bus

Keywords

Travel models (Travel demand), Travel demand, Travel behavior, Trade off analysis, Stop (Public transportation), Scenarios, Ridership, Regression analysis, Regression, Projections, Patronage (Transit ridership), Neural networks, Intracity bus transportation, Fuzzy logic, Fuzzy inference, Forecasting, Delaware, Comparison studies, Case studies, Bus transit, Bus stops, Artificial neural networks, ANNs (Artificial neural networks), Alternatives analysis

Abstract

Fuzzy inference has become a popular approach to modeling systems in which uncertainties associated with human perception and decision are present. Its use for the modeling framework of travel demand prediction is promising. The possibility of applying fuzzy inference to the problem of predicting bus ridership at individual stops is examined. The required properties for such a model are identified, and the mathematics of fuzzy inference is examined. The factors that may cause transit use at individual stops are identified, and their relationships to ridership are modeled using the hierarchically structured fuzzy rule basis. This fuzzy rule-based model is similar to that of the cross-classification approach, with the boundaries of the classes being fuzzy. The artificial neural network and regression methods are used to model the same problem, and the results are compared with those of the fuzzy inference method. The data used for calibration are obtained from a study of actual bus stops in Delaware. Although no conclusion is drawn as to which approach is superior, fuzzy inference provides an alternative to the traditional regression approach in which the phenomenon's causality is complicated yet the behavior of the output must be consistent with commonsense knowledge. In particular, it is suited for problems in which the actual volume to be predicted fluctuates widely. Such problems require strength of logical explanatory power and understanding of the pattern rather than accuracy in fitting the data.

Share

COinS