Citation

Belen R, Taskyaya Temizel T, Kaygisiz. Proc. Road Saf. Four Continents Conf. 2010; 15: 344-356.

Copyright

(Copyright © 2010, Conference Sponsor)

DOI

unavailable

PMID

unavailable

Abstract

The highway accident data provides valuable information such as accident black spot locations and their spatio-temporal change to the experts. With the help of this information, experts may take timely precautions in order to prevent the incidents from happening in the future. There is a challenging detail here. Highway accident data is often manually collected. For example, an officer submits the whereabouts of accidents in terms of latitude and longitude on a form or on a map to the system. Although many systems are designed to verify the legitimacy of the data to be entered, systems can still be prone to user errors. Users may enter illegitimate values which appear to be valid in the system, i.e. a point that is not on a highway (an outlier) or a point on the highway but not correctly entered (an inlier). These erroneous values are called disguised missing data and yet can arise in many different data sets such as health or survey apart from spatial data sets. However, the way they emerge and the way they appear in spatial data sets such as in the highway accident data sets is different and detection is difficult compared to that of survey and health-related data sets. Consequently, their presence can affect the outcome of data analysis tasks severely which may cause decision makers to make inaccurate decisions. Therefore elimination of these values becomes a necessity prior to the data analysis. In this paper, we will explain the common disguised missing value problems in Turkish highway accident data sets. Since the hotspot analysis will be run after data quality is guaranteed, we have focused on disguised missing values on coordinate information. We will describe a framework about how to detect such values automatically. We believe that the results of this study will be of benefit to the data analysts who are working with similar data sets.