Date of Award


Degree Type


Degree Name

Master of Science in Systems Engineering


Mechanical, Industrial and Systems Engineering

First Advisor

Gretchen Macht


Electric Vehicles (EV) sales are experiencing an increasing trend in many industrialized countries [1, 2]. Globally, at the end of 2017, there was an annual increase of one million EVs on the road, totaling to three million EVs on the road [3]. However, despite recent developments and the high potential of Battery Electric Vehicles (BEVs), the market penetration rate of EVs is still very low due to discrepancies between consumer expectations and knowledge, the limited range and long charging times [4, 5]. Recent research demonstrated that there is a significant difference in energy consumption of BEVs between aggressive and non-aggressive driving. This research additionally, provide evidence that the concept of eco-driving for Internal Combustion Engines (ICE) vehicles works well for describing energy efficient Driving Behavior (DB) for BEVs [6].

The goal of this research was to confirm the energy consumption clusters found in the literature, as well as to confirm and expand the clustering methodology executed for determining these clusters. The original literature executed a hierarchical clustering technique utilizing Ward’s algorithm. In addition to verifying the hierarchical clusters, Latent Profile Analysis (LPA), a form of model-based clustering, is then introduced as the new clustering approach to explore alternative clusters through a more diverse clustering approach.

Based on the fact that Dataset 1 (from previous research) and Dataset 2 (from this body of work) were found to be statistically similar, they get merged into a more comprehensive dataset. This research confirmed the two energy consumption clusters (i.e., efficient and inefficient drivers) found in previous research with Dataset 1 using Ward’s method. Given the fact that the clusters were very similar for both Ward’s method and LPA for Dataset 1, these results strongly affirm these previous results regardless of the methodological clustering approach. Clustering Dataset 2 with Ward’s method resulted in three energy consumption clusters as well, providing proof that at least three clusters are significant. LPA for Dataset 2 revealed similar clusters providing evidence that Ward’s method and LPA find similar cluster when the sample size within the clusters is sufficient large.

For the Combined Dataset, excluding the outlier driver 34.1, with a sample size exceeding 50 participants, Ward’s method results in three significant clusters. This strengthens the argument that DB with respect to energy consumption can be clustered into at least three clusters. Expanding the cluster analysis by LPA provides a four and five component model with each equally shaped clusters, grouping drivers in accordance to what is known in the literature about the influence of DB on energy consumption.

This research provides a better understanding of how BEV drivers need to be clustered based on their mean energy consumption per mile and standard deviation. It provides strong evidence that the assumption from previous research, that at least 3 clusters are relevant when analyzing driving behavior with respect to energy consumption, is true. Additionally, further clusters are found on a more comprehensive dataset which go along with the perception of literature that acceleration and speed are main factors for explaining energy consumption of BEV driving behavior.