Loading...

Examining the Factors Influencing the Severity of Motorcycle Accidents: A Case Study in Eight Provinces of Iran to Investigate Geographical Location Differences

Pandamooz, Romina | 2023

11 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 56939 (09)
  4. University: Sharif University of Technology
  5. Department: Civil Engineering
  6. Advisor(s): Kermanshah, Amir Hassan
  7. Abstract:
  8. Among motorized vehicles, motorcycles are recognized as a common mode of transportation in countries with low to moderate income. Iran is also one of the countries where motorcycles are considered a primary means of transportation. However, in recent years, motorcycle accidents have sharply increased, with 30 percent of accidents in Iran involving motorcycles. This highlights the importance and necessity of taking measures to enhance safety and reduce motorcycle accidents in Iran. In this study, data on intercity motorcycle accidents in the provinces of Khorasan Razavi, Isfahan, Qom, Mazandaran, Gilan, Hamedan, Hormozgan, and Ilam were used. The main objective was to identify and compare the best machine learning algorithms in predicting the severity of motorcycle accidents at three levels (property damage only, serious injury, and fatal injury) in eight provinces of Iran. Seven machine learning algorithms, including Decision Tree, Random Forest, K-Nearest Neighbors, Support Vector Machine, Multinomial Naive Bayes, eXtreme Gradient Boosting, and CatBoost, were employed. Then, the study focused on determining the effective factors on the severity of motorcycle accidents using the best-identified algorithm in each province. The results indicated that the eXtreme Gradient Boosting algorithm in four provinces (Khorasan Razavi, Qom, Hamedan, and Ilam) and the CatBoost algorithm in four provinces (Qom, Gilan, Hormozgan, and Ilam) had the highest accuracy in predicting the severity of motorcycle accidents. Ensemble learning algorithms performed better than the other algorithms in predicting accident severity. Gradient-boosted algorithms also demonstrated higher classification ability compared to the bagging approach. However, K-Nearest Neighbors and Multinomial Naive Bayes algorithms had lower accuracy in predicting these events in most provinces. Furthermore, human factors and vehicle defects had the most significant meaningful impact on the severity of motorcycle accidents. Considering the results of this study, a combination of effective regulations with awareness promotion, comprehensive education, and active monitoring is crucial to reduce the severity of motorcycle accidents. Additionally, enforcing strict laws for obtaining standard technical licenses for motorcycle riders can be effective
  9. Keywords:
  10. Crash Severity ; Motorcycle ; Machine Learning ; Road Safety ; Accident Prediction

 Digital Object List

 Bookmark

No TOC