Forecasting Traffic and Freight Demand in order to decide on Expansion

Project Details

Term: 

2019

Students: 

Sangyong Lee, Jaspreet Nayyar, Arun Nakkeeran, Rahul Mohapatra, Debayan Deb, Niladri Mukherjee

University: 

ISB

Presentation: 

Report: 

a. Problem Description

Our client, The Australian Airports Association, appointed us for forecasting the passenger and freight traffic for the month of September 2019. Based on our forecast they intend to take the decision of whether to go ahead with investing a budgeted $1.2 Billion on building new infrastructure (including runways, facilities) and recruiting staff. As per their own business estimates, they have set a threshold of a 40% increase in their YoY traffic for both passenger and freight data to decide for a go/no go on the Investment.

To undertake the project of infrastructure expansion, the Airport Association needs one year for implementation and needs the forecasted traffic for the next year every month, so that the project can be started at an implementable pace. Hence the Business model is based on monthly forecasted Traffic for one year ahead to facilitate the authorities in the decision making.

b. The dataset

To accurately forecast the passenger and freight traffic our team has used government data from the Australian Government; Data source: https://data.gov.au/dataset/ds-dga-d9fbffaa-836f-4f52-80e8- 324249ff269f/details.

We have plotted the dataset against time and the brief analysis of the plot shows a shift in traffic trend post-Sep 2001. To make the data more relevant to the current demand forecast, we have truncated the data from Oct 2001 to Sep 2018 (This truncation was done based on changes in worldwide air transport regulations post 9/11 attacks in the USA).

c. Metrics

Since the business decision is based on the incremental change in the Passenger & Freight traffic, we have used MAPE (which acts as a proxy for the incremental change) to benchmark different models as it provides the percentage change in forecasted traffic vs actual Traffic.

To decide the best model for predicting each data set, The MAPE for both training and validation were compared to look for overfitting and then the model with least values of MAPE has been decided as the best predictor (since the data differences were very small, the decision to go ahead with the model with the minimum MAPE was used to forecast the data).

d. Analysis and Final forecasting method

We have analysed each of 6 datasets- passenger and freight data for cities of Brisbane, Sydney and Melbourne- with multiple methods of prediction (as indicated in technical summary) and each model used a different final method for forecasting based on MAPE data. After a detailed analysis, we have concluded that the decision is a no go for all 3 airports as the increment of traffic in each of the 6 cases is less than the stipulated 40%.

Application Area: