Database Reference
In-Depth Information
mans or animals) heavily affect the quality of the sensor data [31, 73,
23]. This may cause a significant problem with respect to data utiliza-
tion, since applications using erroneous data may yield unsound results.
For example, scientific applications that perform prediction tasks us-
ing observation data obtained from cheap and less-reliable sensors may
produce inaccurate prediction results.
To address this problem, it is essential to detect and correct erroneous
values in sensor data by employing
data cleaning
. The data cleaning task
typically involves complex processing of data [71, 30]. In particular, it
becomes more dicult for sensor data, since true sensor values corre-
sponding to erroneous data values are generally unobservable. This has
ledtoanewapproach-
model-based data cleaning
. In this approach,
the most probable sensor values are inferred using well-established mod-
els, and then anomalies are detected by comparing raw sensor values
with the corresponding inferred sensor values. In the literature there
are a variety of suggestions for model-based approaches for sensor data
cleaning. This section describes the key mechanisms proposed by these
approaches, particularly focusing on the models used in the data cleaning
process.
3.1 Overview of Sensor Data Cleaning System
A system for cleaning sensor data generally consists of four major
components:
user interface, stream processing engine, anomaly detector
,
and
data storage
(refer
Figure 2.5
). In the following, we describe each
component.
user interface
stream processing engine
anomaly detector
sensors
i
2
1
:
t
i
10:2
11:2
:
v
ij
10.1
10.9
:
i
2
1
:
t
i
10:2
11:2
:
v
ij
fixed
10.9
:
raw sensor data
cleaned data
(materialized views)
data storage
Figure 2.5.
Architecture of sensor data cleaning system.
