Skootik – Analysis and prediction of pedestrian flow in a city
Skootik
Sector: ICT
Business Case
A given city has several sensors that quantify the number of people passing along certain streets every 15 minutes. These sensors are distributed in several neighbourhoods both in the centre and on the periphery. The aim is to estimate future pedestrian traffic for the consequent decision making by the public authority regarding public safety, mobility, public services such as maintenance of street furniture and waste collection, strategic plans for the promotion of local business, etc.
Objectives
For each sensor, predict future pedestrian flow. The working calendar (holidays) and cultural calendar (events, festivities, etc.) as well as meteorological data must be taken into account.
Use case
A Holt-Winters type seasonal time series model (exponential smoothing) is proposed. A system is also developed to collect weather data through the API of a weather forecasting platform.
Infrastructure
It is a Cloud-based service, so a server is provided to host the system along with a platform for user interaction with the data and a system for ingesting and preprocessing the client’s data.
Technology
Automatic or deep learning
Data
Private data on pedestrian traffic, and public data on weather, work and cultural calendars.
Resources
Dos personas de backend; una de ellas más centrada en el tratado del dato, en la ingesta y en el preproceso, y otra en la algoritmia y en validar los resultados. Otra persona de frontend para el desarrollo de la plataforma para la visualización de los datos.
Difficulties and learning
The data is highly changeable (new data comes in every 15 minutes), so it is important to have a good system of logs and historical data analysis to be able to replicate errors that have happened in the immediate past.
KPIs (business impact and metrics of the model)
The usual metrics for Machine Learning (MSE, MAD, etc.) were used as metrics. This system has proven to be useful not only for forecasting future mobility, but also for analysing historical mobility and determining trends and quantifying pedestrian surges on specific dates.
Funding
Public funding from the relevant public institution for mobility and smart cities
Collaborators, Partners
City Council
