Geographically Weighted Regression as a Predictive Tool for Station-Level Ridership : The Case of Stockholm

• Main Author: Ounsi, Karim
• Format: Others
• Language: English
• Published: KTH, Transportplanering 2019
• Subjects: Other Engineering and Technologies; Annan teknik
• Online Access: http://urn.kb.se/resolve?urn=urn:nbn:se:kth:diva-259500

This thesis studies a new regression method, Geographically Weighted Regression (GWR)to predict ridership at the station level for future stations. The case study of Stockholm’s blue lineis used as new stations will be built by 2030. This paper is written in English.Historically, linear regression methods, independent of the geographical location of theobservations, was and is still used as the Ordinary Least Square regression method. With the riseof GIS-softwares these last decades, geographically dependent regression can be used and previouspreliminary studies have shown a dependency between ridership and location of the station withinthe network.GWR equations for new stations are determined and used to predict their respectiveridership. GIS-data was collected using Geodata and Traffikverket (Traffic Authority) andridership as well as socio-economic related material for the base year of 2016 was used in order todetermine, first, significant variables from a group of candidate ones (Workers, number of buslines and type of change were chosen) and, second the OLS and GWR equations. Significances ofboth models were compared and the OLS equation was used in order to determine the hypotheticalridership of the new stations if they were present in 2016. GWR equations were then determinedusing these calculated ridership of these new stations. Having all GWR equations (each stationhaving its own equation), ridership was thus predicted for the new stations for 2030 usingassumptions and planned, programmed development around the stations (population, apartment tobe built…) and compared with the official predictions.The results show that the GWR method, generally, overpredicts ridership when comparedto the official predictions. Many reasons can explain this overprediction like the assumptions madewith regards to the number of buses as well as the method followed to calculate the number ofworkers around each station.Three main conclusions were drawn for this case study. One main conclusion, specific forthis study and two other, more general, conclusions were deduced from this study. First, GWR isa good predicting tool for future stations that are close to most currently available stations. Second,GWR is a good predicting method for stations where limited changes in the future environmentwill occur.