13 Smart Learning Using Big and Small Data for Mobile and IOT e-Health
613
The range of self-evaluated stress level is from 0 (“not at all”) to 10 (“extremely
likely”). In addition, exogenous and environmental variables retrieved from public
archives, such as day in a week, daylight time, temperature and precipitation, are
also included. These assessment instruments have been used extensively in the
literature with smartphone for EMA in recent studies [60].
The continuously collected personal health data is one type of the ecological
momentary assessments (EMA), which are used to repeatedly sample individuals’
experience and behavior in real time under their natural environments [61]. The
advantage of collecting EMA data is that it reduces recall bias and allows for a
closer investigation of the process dynamics that influence individuals’ real-world
state and behavior. EMA data collected from mobile and IOT e-health devices can
be used to personalize messages to individuals [28] and to support their self-care
management of certain concerns [62].
Powerful statistical methods have been proposed to model massive mobile health
data and characterize the bidirectional relationship between stress and exercise
behavior. With the advance of mobile and IOT technologies, data can be collected
in real time over a long time horizon. For data collected from an individual, referred
to as “small data,” N-of-1 analytical methods are available to estimate individual
effect and to provide personalized healthcare advices. To handle data collected
from a large group of individuals, referred to as “big data,” actionable learning
methods (inspired by reinforcement learning) are able to learn population-level
patterns and make recommendations to individuals. These methods offer a smart
learning framework to model large-scale and complex mobile and IOT e-health data
and shed light on the diverse and bidirectional nature of the relationship between
stress and exercise behavior in general.
13.2.2 Case Study: N-of-1 Analytical Methods
Individual information can be collected via mobile and IOT e-health technologies
for a long or even indefinite period of time. It is challenging to gain insights
from a massive volume of data to understand how modifiable variables influence
health outcomes of interest. The understanding of the relationship is traditionally
obtained at a population level. However, the environment exposures, individual
characteristics, and modifiable elements that affect the dynamics of the relationship
are usually individual-specific. Therefore, individual-specific analytical approaches
are desirable to design personalized guidance and tailor for individual needs. This
motivates the use of N-of-1 methods instead of the traditional population-based
approach. An N-of-1 trial is a clinical trial in which only a single individual is in
the entire trial and in which random allocation can be used to decide the order of
interventions given to the target individual. N-of-1 analytical methods utilize the
data collected from N-of-1 trials and provide a personalized assessment that leads
to better learning.
613
The range of self-evaluated stress level is from 0 (“not at all”) to 10 (“extremely
likely”). In addition, exogenous and environmental variables retrieved from public
archives, such as day in a week, daylight time, temperature and precipitation, are
also included. These assessment instruments have been used extensively in the
literature with smartphone for EMA in recent studies [60].
The continuously collected personal health data is one type of the ecological
momentary assessments (EMA), which are used to repeatedly sample individuals’
experience and behavior in real time under their natural environments [61]. The
advantage of collecting EMA data is that it reduces recall bias and allows for a
closer investigation of the process dynamics that influence individuals’ real-world
state and behavior. EMA data collected from mobile and IOT e-health devices can
be used to personalize messages to individuals [28] and to support their self-care
management of certain concerns [62].
Powerful statistical methods have been proposed to model massive mobile health
data and characterize the bidirectional relationship between stress and exercise
behavior. With the advance of mobile and IOT technologies, data can be collected
in real time over a long time horizon. For data collected from an individual, referred
to as “small data,” N-of-1 analytical methods are available to estimate individual
effect and to provide personalized healthcare advices. To handle data collected
from a large group of individuals, referred to as “big data,” actionable learning
methods (inspired by reinforcement learning) are able to learn population-level
patterns and make recommendations to individuals. These methods offer a smart
learning framework to model large-scale and complex mobile and IOT e-health data
and shed light on the diverse and bidirectional nature of the relationship between
stress and exercise behavior in general.
13.2.2 Case Study: N-of-1 Analytical Methods
Individual information can be collected via mobile and IOT e-health technologies
for a long or even indefinite period of time. It is challenging to gain insights
from a massive volume of data to understand how modifiable variables influence
health outcomes of interest. The understanding of the relationship is traditionally
obtained at a population level. However, the environment exposures, individual
characteristics, and modifiable elements that affect the dynamics of the relationship
are usually individual-specific. Therefore, individual-specific analytical approaches
are desirable to design personalized guidance and tailor for individual needs. This
motivates the use of N-of-1 methods instead of the traditional population-based
approach. An N-of-1 trial is a clinical trial in which only a single individual is in
the entire trial and in which random allocation can be used to decide the order of
interventions given to the target individual. N-of-1 analytical methods utilize the
data collected from N-of-1 trials and provide a personalized assessment that leads
to better learning.