How Much Data is Enough? A Statistical Approach with Case Study on Longitudinal Driving Behavior

Abstract

Big data has shown its uniquely powerful ability to reveal, model, and understand driver behaviors. The amount of data affects the experiment cost and conclusions in the analysis. Insufficient data may lead to inaccurate models while excessive data waste resources. For projects that cost millions of dollars, it is critical to determine the right amount of data needed. However, how to decide the appropriate amount has not been fully studied in the realm of driver behaviors. This paper systematically investigates this issue to estimate how much naturalistic driving data (NDD) is needed for understanding driver behaviors from a statistical point of view. A general assessment method is proposed using a Gaussian kernel density estimation to catch the underlying characteristics of driver behaviors. We then apply the Kullback-Liebler divergence method to measure the similarity between density functions with differing amounts of NDD. A maxminimum approach is used to compute the appropriate amount of NDD. To validate our proposed method, we investigated the car-following case using NDD collected from the University of Michigan Safety Pilot Model Deployment (SPMD) program. We demonstrate that from a statistical perspective, the proposed approach can provide an appropriate amount of NDD capable of capturing most features of the normal car-following behavior, which is consistent with the experiment settings in many literatures.

An Accelerated Approach to Safely and Efficiently Test Pre-Production Autonomous Vehicles on Public Streets

“…The predominant reason for this is the low exposure of the safety-critical events, such as crashes, that makes the training dataset sufficient for making meaningful statistical inference to be extremely large to the extent that collecting it under Naturalistic Field-Operational Test (N-FOT), i.e. on-road deployment, becomes time-consuming, risky, and costly [2], [3], [4]. Google cars, Waymo, for instance, have logged over 3.5 million miles from four states in the USA: Washington, California, Arizona, and Texas [5].…”

A Multi-Vehicle Trajectories Generator to Simulate Vehicle-to-Vehicle Encountering Scenarios

“…We used driving encounter data collected by the University of Michigan Transportation Research Institute (UMTRI) [4] from about 3,500 vehicles equipped with on-board GPS. The latitude and longitude data was recorded by the GPS device to represent vehicle positions.…”

Learning and Inferring a Driver's Braking Action in Car-Following Scenarios

“…In addition, if the relative range was less than 10 m, the event ended, ensuring that no Stop-&-Go case was included. • The duration for a singular car-following event should be larger than 50 s. • Any driver with less than 500 car-following events was eliminated, which was able to guarantee the collected data were enough to capture the underlying driving styles [46]. Based on the above criteria, 49 drivers were selected out from 56 drivers.…”