Forecast daily cab demand by origination of bookings

Project Details




Ankit Kansal, Garrett Butler, Rahul Gupta, Shruti Jain, Vikram Deshpande





1 Problem description
After visualizing the data in the first cut, we picked up the issue to forecast regional level demand. In the Indian context, where radio cabs have flooded the market, it has become very necessary to plan and forecast the demand at a regional level. Yourcabs works on taxi aggregator model. Hence, to devise a real time schedule for vendors based on the regions that they operate is very critical. This will help not only to manage capacity allocation but also to develop new vendor relationships.

2 Brief description of the data, its source, key characteristics, and chart(s)
The data given to us contains the historical demand numbers at a latitude and longitude level both at origin and destination. We have picked up the demand origins in the light of business issue mentioned above. Further, the data contains the mobile, phone and online booking as three different demand patterns. We have strived to forecast the aggregate demand by combining each of the three series independently. Figure 3 provides the visualizations of the series used.

3 High level description of the final forecasting method and performance on meaningful metrics (compared to benchmark)
We used a wide variety of methods ranging from Naïve to Neural net and ensembles to arrive at the final forecasts. We started with the Naïve method for each of the series and used it as a benchmark for any future models. Our final recommendation uses a mix of the following based on the patters for each of the series: Regression +AR (1) [Total Demand, Mobile bookings], Ensemble (Lag2+Neural Net) [Phone bookings], Holt Winters [Online bookings], Ensemble [demand ratio for region 1].
To arrive on a conclusion for the best methods, we benchmarked each method’s output from the naïve forecast, MAPE figures, ACF plot for residuals, and visualization of residuals.

4 Conclusions and recommendations
The daily demand when forecasted using the aggregate of the bookings from the three channels performs better than the forecast of the total demand. This demand when used with the forecasted proportion of the demand from the top-booking region can give us the estimated demand from that region. In view of the increasing overall trend of demand, yourcabs can use this information to manage vendor relationships and capacity allocation for better serving its customers.

Application Area: