Database Reference
In-Depth Information
In this approach, the user registers SQL-like queries that define
constraints over the sensor values. Sensor values are marked as
outliers when these constraints are violated. In addition to these
methods, we also discuss many other data cleaning approaches [31,
73, 23, 21, 52, 65]
Query Processing:
Obtaining desired answers, by processing
queries is another important aspect in sensor data management.
In this chapter, we discuss the most significant model-based tech-
niques for query processing. One of the objectives of these tech-
niques is to process queries by accessing or generating minimal
amount of data [64, 5]. Model-based methods that access/generate
minimal data, and also handle missing values in data, use models
for creating an abstraction layer over the sensor network [18, 33].
Other approaches model the sensor values by a hidden Markov
model (HMM), associating state variables to the sensor values. It,
then, becomes ecient to process queries over the state variables,
which are less in number as compared to the sensor values [5].
Furthermore, there are approaches that use dynamic probabilistic
models (DPMs) for modeling spatio-temporal evolution of the sen-
sor data [33, 29]. In these approaches, the estimated DPMs are
used for query processing.
Data Compression:
It is well-known that large quantity of sen-
sor data is being generated by every hour. Therefore, eliminating
redundancy by compressing sensor data for various purposes (like,
storage, query processing, etc.) becomes one of the most challeng-
ing tasks. Model-based sensor data compression proposes a large
number of techniques, mainly from the signal processing literature,
for this task [1, 72, 22, 53, 7]. Many approaches assume that the
user provides an accuracy bound, and based on this bound the sen-
sor data is approximated, resulting in compressed representations
of the data [24]. A large number of other techniques exploit the
fact that sensor data is often correlated; thus, this correlation can
be used for approximating one data stream with another [24, 67,
49, 3].
This chapter is organized as follows. In Section 2, we define the pre-
liminaries that are assumed in the rest of the chapter, followed by a
discussion of important techniques for sensor data acquisition. In Sec-
tion 3, we survey model-based sensor data cleaning techniques, both
on-line and archival. Model-based query processing techniques are dis-
cussed in Section 4. In Section 5, model-based compression techniques
