National Technical University of Athens (NTUA)
Benchmarking driving efficiency using data science techniques applied on large - scale smartphone data
Abstract
dc:descriptionThe main objective of this PhD is to provide a methodological approach for benchmarking driving efficiency in terms of safety on a trip and driver basis using data science techniques. The methodological approach is based on the definition of a safety efficiency index based on the Data Envelopment Analysis (DEA) theory and is related to macroscopic behavioral driving characteristics such as the number of sudden accelerations / decelerations, mobile phone usage time, and time exceeding the speed limit. In this dissertation, machine learning models are also developed to identify the different driving profiles that exist based on the temporal evolution of driving efficiency. The proposed methodological approach is applied on a real-world driving dataset collected from smartphones, which is analyzed using statistical methods to determine the amount of driving data to be used in the analysis. The results show that the optimized convex hull - DEA algorithm gives the exact solution in significantly less time than the classic DEA approaches. Furthermore, the methodology allows for the identification of the least efficient trips in a database as well as the efficient level of driving metrics of a trip to make it more efficient in terms of safety. Further clustering of the drivers based on the temporal evolution of driving performance leads to the identification of three main driver groups, the typical driver, the unstable driver and the less dangerous driver. Results indicate that having a prior knowledge on user's accident history solely affects the composition of the second cluster of the most volatile drivers, which incorporates drivers that are less efficient and unstable in terms of safety. It is shown that mobile phone use is not a critical factor in determining the safety efficiency of a driver since slight differences are found with regards to this characteristic between drivers of different efficiency classes. Furthermore, it is shown that a different driving data sampling is required for each a) road type, b) driving characteristic and c) driving aggressiveness level to collect enough data to obtain a clear picture of a driver’s behaviour and perform DE analysis. Results could be exploited to provide personalized feedback to drivers on their total driving efficiency and its evolution in order to improve and reduce accident risk.
Degree
thesis:*- Grantor dc:publisher
- National Technical University of Athens (NTUA)
- Year dc:date
- 2018
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Tselentis, Dimitrios
- Τσελέντης, Δημήτριος
Subjects
dc:subject × 16- Οδηγική συμπεριφορά
- Δεδομένα smartphone
- Συγκριτική ανάλυση
- Επιστήμη δεδομένων
- Οδηγική επίδοση
- Driving behavior
- Smartphone data
- Benchmarking
- Data science
- Driving Efficiency
- Επιστήμες Μηχανικού και Τεχνολογία
- Επιστήμη Πολιτικού Μηχανικού
- Επιστήμη Ηλεκτρολόγου Μηχανικού, Ηλεκτρονικού Μηχανικού, Μηχανικού Η/Υ
- Engineering and Technology
- Civil Engineering
- Electrical Engineering, Electronic Engineering, Information Engineering
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/44854