Database Reference
In-Depth Information
Table 14.1.
Sensor data mining challenges at each stage of the data mining process
(I) Acquisition
(II) Pre-processing
lack of data standards
data formatting
lack of data protocols
data normalization
data privacy
data synchronization
(III) Transformation
(IV) Modeling
physiological feature extraction
sequential mining
feature time scales
distributed mining
unstructured data
privacy preserving modeling
obtaining ground truth
exploration-exploitation trade-offs
(V) Evaluation and Interpretation
Model expressiveness
Process and data provenance
data models and protocols to externalize sensed signals. In healthcare,
standard bodies like HL7 [9] and the Continua Health Alliance [10]
address data modeling issues while several IEEE standard protocols ad-
dress device interoperability issues [11]. However, there is a lack of
incentives for sensor data manufacturers to adhere to these standards.
With this lack of adherence to standards, mining medical sensor data
across multiple data sources involves several non trivial engineering chal-
lenges, and the design of custom solutions specific to each sensor data
mining application.
Another key challenge in the acquisition process is related to the pro-
tection of user privacy. In United States, the Health Insurance Porta-
bility and Accountability Act (HIPAA) defines regulations on accesses
to health data. By law, data mining applications that leverage this
data must comply with these regulations. Data de-identification and
de-anonymization techniques are often required to comply with HIPAA.
Privacy preserving data mining techniques [12], [13] may also be used to
extract information form sensor data while preserving the anonymity of
the data.
2.2.2 Pre-processing Challenges.
Data in the real world is
inherently noisy. The pre-processing stage needs to address this problem
