Database Reference
In-Depth Information
Examples include motion detection sensors, audio and video sen-
sors, temperature sensors, weather sensors, etc.
2.2 Challenges in Mining Medical Informatics
Sensor Data
As with standard data mining procedures [7] [8], healthcare mining is
typically performed in five stages:
(I) Data Acquisition
: This includes operations involved in collect-
ing data from external sensor data sources.
(II) Data Pre-processing
: This includes operations applied to the
data to prepare it for further analysis. Typical pre-processing op-
erations include data cleaning to filter out noisy data elements,
data interpolation to cope with missing values, data normalization
to cope with heterogeneous sources, temporal alignment, and data
formatting.
(III) Data Transformation
: The includes operations for represent-
ing the data appropriately and selecting specific features from this
representation. This stage is often called feature extraction and
selection.
(IV) Modeling
: This stage, also called
mining
applies knowledge
discovery algorithms to identify patterns in the data. Modeling
problems can be classified into six broad categories: (1) anomaly
detection to identify statistically deviant data, (2) association rules
to find dependencies and correlations in the data, (3) clustering
models to group data elements according to various notions of
similarity, (4) classification models to group data elements into pre-
defined classes, (5) regression models to fit mathematical functions
to data and (6) summarization models to summarize or compress
data into interesting pieces of information.
(V) Evaluation
: This stage includes operations for evaluation and
interpretation of the results of the modeling process.
There are several analytical challenges associated with each of these
stages - specific to healthcare mining - that are listed in Table 2.2.
We present these analytical challenges in more detail in the rest of
this section.
2.2.1 Acquisition Challenges.
Despite several standardiza-
tion efforts, medical sensor manufacturers tend to design proprietary
