The World Bank says Twitter could help end Nairobi road accidents
If you follow @Ma3route on Twitter, you are more likely to avoid traffic snarl-ups by getting live tweets reporting road carnage incidents during your commute.
If the authorities at the Traffic police would bother to do the same, they would discover that 50 per cent of road crashes reported by the Twitter account over the last eight years have happened in just 133 locations in the city.
The World Bank has done a study of Twitter accident reports which could revolutionize how traffic police work is done and help reduce the number of accidents and wanton loss of life.
“When looking at the 7.5 years of crowdsourced data for the city of Nairobi, the number of crash clusters does not grow linearly, implying that the locations where crashes occur and are reported on Twitter are consistent across years. We see the concentration of crashes, even more, when we note that only 9 per cent of crash clusters (133 of1,375) represent 50 per cent of the crashes reported,” the World Bank said.
“Only 14 per cent of crash locations have only a single crash, and there are 443 crash clusters with 10 or more crashes,” the report, Applying Machine Learning and Geolocation Techniques to Social Media Data (Twitter) to develop a Resource for Urban Planning read.
Read also: Banking on tech solutions to unclog Nairobi Traffic Jams
Even with coronavirus pandemic which saw Nairobi, Mombasa and Mandera counties being put on lockdown and dusk to dawn curfew imposed the number of people who died from accidents still shot up 9.4 per cent to 3,663 people by December last year according to data by National Transport and Licensing Authority (NTSA).
World Bank says analyzing the crash data produced using their algorithm and focusing on the truth dataset within the city limits of Nairobi, reveals trends that can help city authorities make interventions and reduce road crashes.
The data shows that all crashes from July 2017 to July 2018 can be found in 435 clusters, each with a maximum diameter of 300 m.
Of these clusters, 67 per cent have two or more crashes and there are 56 clusters with 10 or more crashes. Additionally, 66 crash clusters represent over 50 per cent of all the crashes.
The study tracked twitter for words such as ajali, accident, crash, roll and gathered 52,228 tweets as crash-related out of a universe of 874,588 tweets during 2012 – 2020.
The researches geolocated 32,991 time-stamped crash tweets from August 2012 to July 2020 and clustered them into 22,872 unique geolocated crashes.
They found that 63 per cent of tweets contain enough information in order to be geolocated.
Assuming the same proportion of tweets contain enough information to be geolocated in the full dataset, they would expect 32,903 tweets with enough location information
They found that of the 73 crash-related tweets physically verified, 92 per cent correctly corresponded to a crash near the estimated location; 32.8 per cent witnessed the crash scene, 57.5 per cent did not see the crash but were told by a bystander that a crash occurred and was recently cleared, and 1.4 per cent reported that the crash did not occur at the specified location but nearby.
