Experimental study on driving scenarios and driver behaviours in Malaysia by using machine learning techniques

<p>The increasing number in annual road fatalities has caused a major challenge in many</p><p>countries. Minimising fatalities and improving safety are the top priorities of different</p><p>countries. This study aimed to analyse d...

全面介紹

Saved in:
書目詳細資料
主要作者: Garfan, Salem Abdullah Salem
格式: thesis
語言:eng
出版: 2021
主題:
在線閱讀:https://ir.upsi.edu.my/detailsg.php?det=7052
標簽: 添加標簽
沒有標簽, 成為第一個標記此記錄!
實物特徵
總結:<p>The increasing number in annual road fatalities has caused a major challenge in many</p><p>countries. Minimising fatalities and improving safety are the top priorities of different</p><p>countries. This study aimed to analyse driver behaviours in Malaysia and the impacts</p><p>of practising eco-driving to improve safety, reduce fuel consumption and green gas</p><p>emission by using smartphone sensors and OBD2 (ELM327) adapter based on event</p><p>thresholds and machine learning algorithms. In the experimental study, 30 drivers had</p><p>participated, which were 17 novice drivers (7 males and 10 females) and 13 experienced</p><p>drivers (8 males and 5 females). A Honda Civic 2019 car was used in the experiment.</p><p>A specific route was selected for all drivers, which consisted of two types of road</p><p>(highway and urban), with a total distance of 20.6 km. The analysis of driving behaviour</p><p>was based on threshold events and machine learning algorithms. This was to classify</p><p>the different driving scenarios. In the drivers profiling, driving behaviour was</p><p>categorised into three driving behaviours, such as safe, normal, and aggressive driving.</p><p>Random Forest model was selected for the classification after being compared to other</p><p>different machine learning algorithms (Decision Tree, Support Vector Machine, KNearest</p><p>Neighbour, and Nave Bayes models). The results of this experiment showed</p><p>that a remarkable reduction in terms of fuel consumption and CO2 emission of up to</p><p>30% less was achieved when participants followed the eco driving techniques.</p><p>Moreover, aggressive events were notably reduced in eco driving as compared to</p><p>normal driving. Furthermore, the selected machine learning model was able to</p><p>differentiate and classify different driving scenarios with high classification accuracy</p><p>of up to 100 %, such as identifying male and female drivers, novice and experienced</p><p>drivers, and driving in the highway or city.</p>