Seoul Institute Logo

You are here


A Study on Big Data Based Method of Classification and Demand Estimation for Public Parking Lot

Dong-Guen KimㆍBum Sik KimㆍYong-Hoon KwonㆍSang Yeon HongㆍYoung- Hyun ShinㆍHyo-Sub SimㆍKwiwon Park

The Seoul Public Investment Management Service is continuously researching ways to estimate the demand for public parking lots. However, estimating resident demand using the number of parking spaces and the number of registered vehicles has limitations. Therefore, it is necessary to develop an estimation methodology for total parking demand including part-time parking. This study collected data expected to affect the demand for public parking lots such as parking performance, commercial data, zoning data and distribution of public transportation facilities and competitive parking lots. In this study, a methodology of public parking lot classification was developed using time-series cluster analysis and the methodology of demand estimation of public parking lot using machine learning.

As a result of performing time-series cluster analysis, the result of Euclidean K-means cluster analysis with the number of clusters set to 4 showed the best explanatory power. The following clusters were derived from Euclidean K-means analysis: commercial area type with a small occupancy amplitude and low occupancy on weekends compared to weekdays; commercial + housing area type with a large occupancy amplitude showing similar occupancy rates to weekdays even on weekends; and night time zone. It was found that a residential area type with a high occupancy rate did not show a large difference between weekdays and weekends. It was also found that a transfer parking lot type had a large occupancy amplitude and a low occupancy rate on weekends compared to weekdays. As a result of verifying the result of classification of types by parking performance through the actual area ratio of use area, types of parking lot were correlated with zoning.

In addition, the demand for new public parking lots was estimated through ensemble method with quantile regression analysis based on collected big data. The ensemble method included the voting method and the stacking method. In the quantile regression analysis, a Gradient Boosting Regressor (GBR) model was used. As a result of the analysis, it was confirmed that it was possible to estimate the demand for new public parking lots through ensemble method and quantile regression analysis.

However, limited period and amount of parking performance data used were limitations of this study. It is necessary to periodically investigate the use of public parking lots and to prepare a plan to manage and operate public parking lots using parking performance. Further studies need to perform parking lot type prediction and demand estimation.