Classification of driver behaviours using machine learning
<p>According to the Malaysian Institute of Road Safety Research (MIROS), over</p><p>500,000 car accidents occurred in 2016, making cars an unsafe means of</p><p>transportation. This research aimed to collect driver behaviour-relat...
Saved in:
| Main Author: | |
|---|---|
| Format: | thesis |
| Language: | eng |
| Published: |
2021
|
| Subjects: | |
| Online Access: | https://ir.upsi.edu.my/detailsg.php?det=7127 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | <p>According to the Malaysian Institute of Road Safety Research (MIROS), over</p><p>500,000 car accidents occurred in 2016, making cars an unsafe means of</p><p>transportation. This research aimed to collect driver behaviour-related data for</p><p>Malaysian drivers to provide useful insights for Malaysian driving profile and to</p><p>modulate machine learning for classification tasks. Twenty-one drivers (11 male and</p><p>10 female) were studied and compared for their driving style in Lebuhraya Behrang</p><p>Stesen-tg malim (11 km per driver). Drivers were asked to drive naturally while</p><p>considering their safety. Two analysis techniques were utilized (i.e. Statistical and</p><p>Machine Learning-Based). Different conclusions were drawn from each analysis. The</p><p>number of driving events for each driver was calculated (i.e. aggressive, normal and</p><p>safe) and statistical tests (i.e. Mean, Standard Deviation, Correlation analysis, Oneway</p><p>ANOVA and T-test) presented significant differences between each driver from</p><p>the same gender versus their peers from the opposite gender. The statistics were</p><p>presented per driver, his/her group and a comparison with their peers. For a driver to</p><p>be considered as aggressive or normal, a challenge was presented because no</p><p>identification measure existed (i.e. threshold for driving event number to be</p><p>considered aggressive or normal). However, each driving event was identified based</p><p>on literature. Finally, it was determined that classifying drivers was possible through</p><p>their gender but not based on their aggressiveness level. One R Machine learning</p><p>classifier presented good accuracy at 95.24 % in comparison with j48DecisionTree,</p><p>Naive Bayes, One R, and SMO-SVM. The implications of the findings of this study</p><p>suggest male and female drivers tend to drive aggressively. A reason for such</p><p>mortality can be because of the cadence of front-end car accidents, which is a clear</p><p>outcome of aggressive driving behaviour (i.e. speeding, braking, etc.). Identifying</p><p>such behaviour using ML will save lives domestically and internationally</p> |
|---|