| VI. RESEARCH WORK DONE IN ACADEMIC INSTITUTIONS RELATED TO THESIS WORK
D. TRAFFIC AND TRANSPORTATION
|4. Bus Travel Time Prediction Using Global Positioning Systems (GPS) Data|
|Date of Start|
|Scope and Objectives|
The scope of this work is on developing a travel time prediction model using GPS data. A case study route of Metropolitan Transport Corporation (MTC) in Chennai is considered for this purpose. The specific objectives are to:
|Type of Study - Field Extensive GPS data were collected on MTC Bus Route No. 21G (Parrys –Tambaram) in Chennai city. Data was collected on three probe buses. Data was extracted in the required format from the raw GPS data. Multiple Linear Regression (MLR) models were developed to predict the travel times to bus stops. Models were evaluated using a separate validation data set. Mean Absolute Percentage Error (MAPE) was used as a measure of closeness between the observed and predicted values.|
|Findings and Conclusions|
|(i) Preliminary data analysis revealed that similar traffic conditions prevail over the route during the peak hours on all weekdays. Thus, Multiple Linear Regression (MLR) models which do well in such recurrent traffic conditions were developed. Of the nine MLR models developed, five models had R-Square values of more than 0.89, indicating good fits.
(ii) Variables like ‘Remaining Number of Bus Stops (BSij)’ and ‘Intersection Delay (IDij)’ were found to be statistically insignificant.
(iii) The success rate of the model is high, with the best model having a low Mean Absolute Percentage Error of 9.0.
(iv) It was observed from the model equation that variables like ‘distance remaining (in terms of six lane, four lane and two lane) from the current bus stop to the target bus stop’ and bus stop dwell times significantly affect the bus travel time.
(v) It was also noted that when the distance was classified in terms of number of lanes, the MAPE reduced from 13.0 to 9.0.
(vi) The use of data from probe buses helped improve the performance of the models.