Database Reference
In-Depth Information
Hence, sensor mining in healthcare requires the use of emerging stream
processing system technology in conjunction with database and data
warehousing technologies. Stream processing systems are designed to
cope with large amounts of real-time data, and their programming mod-
els are geared towards the analysis of structured and unstructured sensor
data. They are also time sensitive and analyze data within small latency
bounds.
Figure 14.2
presents an extended architecture for sensor data
mining that illustrates this integration. The rationale behind this archi-
tecture is to use a stream processing system for the real-time analysis of
sensor data, including the pre-processing and transformation stages of
the analytical data mining process. The sensor data acquisition is per-
formed by a layer of software that interfaces with sensors and feeds into
the stream processing system. The results of the transformation stage
may be persisted in a data warehouse for oine modeling with machine
learning techniques. The resulting models may be interpreted by an-
alysts and redeployed on the stream processing platform for real-time
scoring. In some cases, online learning algorithms may be implemented
on the stream processing system. This integration of stream processing
with data warehousing technologies creates a powerful architecture that
addresses the system challenges outlined above.
3. Sensor Data Mining Applications
As illustrated in
Figure 14.1
, applications of sensor data mining tech-
nologies in healthcare can be classified in two major groups: clinical and
non-clinical applications. In the former group, the main data sources
subjected to the mining process are direct observations of patient physi-
ological states. In these cases, data mining is often applied to these data
sources to build patient models for diagnosis and prognosis. Clinical
applications are typically found inside medical institutions. In contrast,
non-clinical applications have a broader scope and do not limit them-
selves to the mining of patient physiological data. Such applications may
be found both inside and outside of medical institutions. In the rest of
this chapter, we survey the literature in both of these classes.
3.1 Clinical Healthcare Applications
Systems supporting clinical applications of data mining technologies
in healthcare are often called Clinical Decision Support Systems (CDSS).
CDSSs have been used in both in-patient and out-patient scenarios
1
.
1
In-patient scenarios refer to scenarios for patients that are hospitalized for more than
24 hours. Out-patient scenarios refer to the rest of the clinical use-cases.
