Applying machine learning and geolocation techniques to social media data (Twitter) to develop a resource for urban...

Area of Science:

• Geographic Information Systems (GIS)
• Machine Learning Applications
• Public Health Informatics

Background:

• Vital statistics and essential data are difficult to obtain in many parts of the world, hindering development efforts.
• Private companies possess valuable data, but accessibility issues limit its use for public good initiatives like poverty tracking or disease reduction.
• Road traffic crashes are a leading cause of mortality for young people, yet crash data is scarce in resource-poor regions, impeding safety interventions.

Purpose of the Study:

• To investigate the feasibility of using openly available Twitter data to generate crucial road traffic crash location data for urban planning.
• To create a novel dataset of geolocated traffic crashes in Nairobi, Kenya, to support road safety initiatives.
• To demonstrate a methodology for transforming social media data into actionable insights for urban development in resource-limited settings.

Main Methods:

• Scraping 874,588 traffic-related tweets from Nairobi, Kenya, between 2012 and 2020.
• Applying a machine learning model to identify crash occurrences and an enhanced geoparsing algorithm for location identification.
• Utilizing spatial clustering to pinpoint high-crash-frequency road segments and real-time verification with a motorcycle delivery service.

Main Results:

• Successfully geolocated 32,991 crash reports from Twitter, identifying 22,872 unique crashes.
• Achieved 92% accuracy in real-time crash verification, demonstrating the reliability of the Twitter data.
• Produced the first geolocated crash dataset and map for Nairobi, revealing that less than 1% of the road network accounts for 50% of identified crashes.

Conclusions:

• Twitter data can be effectively repurposed to generate essential road safety data in resource-poor environments where traditional data collection is challenging.
• The developed methodology offers a scalable approach for creating other critical datasets for urban planning and development using social media.
• The findings provide urban planners with data-driven insights to prioritize and implement targeted road safety improvements, maximizing the impact of limited resources.