Volume 13, Issue 2 (2013)                   MCEJ 2013, 13(2): 149-159 | Back to browse issues page

XML Persian Abstract Print

1- Tarbiat Modares University
2- -----
Abstract:   (6576 Views)
  Abstract: Heavy transport costs have lead researchers in the field towards more efficient techniques to reduce peak period congestion. One such technique is telecommuting which is planned to, in line with the most travel demand management techniques, to decrease motorized trips during commuting hours. By allowing employees to work from home or a center near home equipped with telecommunication technologies, telecommuting causes an increase in efficiency of the urban transport system and a decrease in fuel consumption costs, air pollution costs and the need to invest infrastructures. Identification of the actual demand for telecommuting is a prime to analyzing the potential consequences of telecommuting in mobility improvement, congestion reduction, and energy conservation. Considering the vast socio-economic dimensions of this technique, the present paper intends to model telecommuting demand for the metropolitan of Tehran, Iran, by employing the artificial neural network (ANN) approach. ANNs are applied as a modeling tool for the complex systems of recognition and prediction, inspired by the interconnectivity of the human nervous system. ANN simulates adaptive interaction between processing elements in parallel architecture. A multi-layer perceptron model using error back propagation is deployed to predict the suitable number of weekdays telecommuting for each employment category. Using the data from an interview-filled questionnaire, designed for this purpose, various structures of ANN models were calibrated based on 80 percent of a 676 size sample. The remaining 20 percent of the preference data was preserved to assess the prediction strength of the model as it encounters unforeseen cases. Four endogenous inputs that inferred from organizational characteristics of employees arrayed the neural network model. Due to unordered nominal values of independent variables, ANN was determined to be an appropriate approach to recognize the telecommuting suitability pattern. The proposed neural network is composed of 21 neurons in 3 layers with tan-sigmoid, log-sigmoid, and linear transfer functions in the corresponding hidden and output layers. Results of the proposed model with 171 unknown parameters, converging after 1800 iterations, indicated a fair capability to replicate observations, such that mean square error, coefficient of determination, and percent correct criteria for the test set equaled, respectively, 1.177, 0.19 and 39 percent. The ANN model successfully estimated the stated quantities of telecommuting days per week, within a range of one day error, to 86 percent and 84 percent correct for the train and test sets, respectively. The evaluation results of train and test subsets are relatively close, which indicates a low generalization error, meanwhile demonstrates the reliability of ANNs to forecast the telecommuting demand.    
Full-Text [PDF 473 kb]   (5363 Downloads)    

Received: 2013/08/19 | Accepted: 2013/04/21 | Published: 2013/08/19

Rights and permissions
Creative Commons License This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.