Authors:
Ziegelbecker, Thomas
Abstract:
Today we produce and capture data at almost each and every step. In many cases, this data is imperfect, due to various defects such as sensor variability, errors in measurement, or by human error. Analysts and decision makers unknowingly base their decisions on such imperfect data, which often leads to poor decisions and high costs. One way to address this problem is to visualize data quality problems to make decision makers more aware of them. Despite existing literature proving that data quality visualization improves decision-making, only li (...)
Today we produce and capture data at almost each and every step. In many cases, this data is imperfect, due to various defects such as sensor variability, errors in measurement, or by human error. Analysts and decision makers unknowingly base their decisions on such imperfect data, which often leads to poor decisions and high costs. One way to address this problem is to visualize data quality problems to make decision makers more aware of them. Despite existing literature proving that data quality visualization improves decision-making, only little research has been conducted in the field of univariate and multivariate data quality visualization. Therefore, the focus of this work will be on incorporating data quality visualization into the data exploration process, where the main contribution is to provide a novel approach for visualizing data quality problems of multivariate time-oriented data in both, overview and detail. For this purpose, a particular domain problem from the drilling industry will be used. The data itself is provided from multiple sensors that transmit time-stamped raw drilling-data, which contains data quality problems such as missing values, invalid values and outliers. In this work I examine existing data quality visualizations for multivariate time-oriented data. Based on this literature research I develop and discuss several design options in overview and detail for visualizing the data quality problems identified in combination with the domain problem. In a subsequent step I implement selected design approaches in a prototype and evaluate them in the context of expert interview sessions. The results of these session are then reported and discussed, providing further rationales for the design choices made. In addition, the results also provide arguments for specific interaction techniques (i.e., combined interactive views) as well as they offer insights into algorithms and technologies used. Overall, the results give conclusions for selecting data quality visualization approaches and make suggestions for further research areas such as the aggregation algorithms for data quality problems.
(Read More)