Severity classification of Finnish motorcycle accidents with machine learning methods
Tuominen, Tuuli (2024)
2024
Tietojenkäsittelyopin maisteriohjelma - Master's Programme in Computer Science
Informaatioteknologian ja viestinnän tiedekunta - Faculty of Information Technology and Communication Sciences
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Hyväksymispäivämäärä
2024-05-21
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-202405175990
https://urn.fi/URN:NBN:fi:tuni-202405175990
Tiivistelmä
Traffic accident fatalities among vulnerable road users, such as motorcyclists, are a global burden requiring effective and evidence-based interventions. Currently the use of machine learning is becoming more common in traffic accident modeling, offering new insights into the causes and risk factors of severe accidents.
The purpose of this thesis was to analyze important factors affecting the severity of motorcycle accidents and to compare the performance of machine learning models in the task of motorcycle accident severity classification. A Finnish motorcycle accident dataset covering years 2005–2021 was acquired from Finnish Transport Infrastructure Agency, and K-nearest neighbor classifier, support vector machine and random forest were trained on the dataset. Additionally, the variable importance values were estimated using ReliefF algorithm and out-of-bag permutation of the random forest.
Of the three models, random forest achieved the highest performance in all six performance measures computed. However, the model’s performance was observed to be significantly lower compared to previous research, potentially caused by the lack of important features in the dataset. Variable importance estimates of both methods indicated the type of the accident to be important, as well as the geographical location and whether the accident occurred on dual-carriageway road. Potentially interesting variables that were identified important by one of the two methods included weather, day of the week, motorcycle age and motorcycle weight.
The purpose of this thesis was to analyze important factors affecting the severity of motorcycle accidents and to compare the performance of machine learning models in the task of motorcycle accident severity classification. A Finnish motorcycle accident dataset covering years 2005–2021 was acquired from Finnish Transport Infrastructure Agency, and K-nearest neighbor classifier, support vector machine and random forest were trained on the dataset. Additionally, the variable importance values were estimated using ReliefF algorithm and out-of-bag permutation of the random forest.
Of the three models, random forest achieved the highest performance in all six performance measures computed. However, the model’s performance was observed to be significantly lower compared to previous research, potentially caused by the lack of important features in the dataset. Variable importance estimates of both methods indicated the type of the accident to be important, as well as the geographical location and whether the accident occurred on dual-carriageway road. Potentially interesting variables that were identified important by one of the two methods included weather, day of the week, motorcycle age and motorcycle weight.