DATE : April 25, 2017
TIME : 3 PM
VENUE : ESB 244
SPEAKER : Ravina Gelda (EE14S045)
GUIDE : Dr. Krishna Jagannathan
GTC Members :
Dr.Gaurav Raina (M)
Dr.Gitakrishnan Ramadurai (CE)
In India, there has been a dramatic increase in the popularity for mobile application based taxi hailing services. A key concern for such taxi services is the supply-demand mismatch problem. The capability to forecast supply and demand accurately in future provides valuable decision support to reduce the supply-demand mismatch.
In this talk, we deal with the problem of supply forecasting in the context of mobile application based taxi hailing services. We propose a forecasting methodology which incorporates information from factors which influence supply, such as driver incentive schemes, holidays, hour of the day and the day of the week. We forecast supply in two different scenarios. Firstly, the supply is forecast for a fixed rectangular (6-level geohashes) spatial partitioning of the city. Using the forecasting methodology proposed along with the multilevel clustering technique to deal with data scarcity, we were able to predict supply with almost 91% accuracy for the heavily used 6-level geohashes. In second scenario, we partition the city into Voronoi regions based on the demand density. Next, we identify optimal temporal resolution for forecasting supply in heavily used Voronoi regions. Using the proposed forecasting technique, we were able to forecast supply with almost 90% accuracy for the heavily used Voronoi regions. Finally, we compare these two scenarios of spatial partitioning of the city for supply forecasting and the results show that Voronoi division based on demand density provides comparable accuracy of forecasts with less computational cost.
ALL ARE CORDIALLY INVITED.