The document discusses using Internet of Things (IoT) and predictive analytics for assisted living. Sensors would be installed around a home to collect data on daily movements over time to determine a "norm". Outlier detection could identify movements outside of this norm. Predictive behavior analysis using the sensor data could determine probabilities of falls or changes in care levels. Visualization and machine learning techniques would analyze the data for patterns to gain insights about households and their activity patterns. This could provide assistance for elderly care by sounding alarms if movements fall outside normal patterns.

1. Internet of Things (IoT)
Application of Predictive Analytics to
Assisted Living
September 19, 2017

2. Internet of Things Analytics
Store data Process data
Apply Predictive Analytics
The Underlying Idea
Collect data
Page 2
Better Value Smarter decisions Better insights
RESULTS

3. Internet of Things Analytics
The use cases for such a framework are endless– from managing appliances within a
household, to managing resources on a greater scale in offices or hospitals, to
even looking at a more city-wide level for environmental or traffic management.
A generalized framework can be later customized according to the use-case
Business Value
Page 3

4. Internet of Things Analytics: Structure
Device 3
Device 2
Device 1
Hub/
Cloud Process signals Machine Learning
Predictive General
Data
Intelligence
Communication
Physical
(Sensory)
Page 4
Device 3 Predictive
Maintenance
General
Governance
Analyze
Act
New Optimized
Systems

5. Assisted Living: The Idea
How It works
Install various sensors around the house to track daily movements. Data
collected over a certain time period can be used to determine the “norm.” An
outlier detection system can be created for data deterring from this norm.
Business Value
For elderly care, equipping the home with numerous sensors (e.g., motion,
temperature, heating) to provide assistance. Predictive behavior analysis for
Page 5
temperature, heating) to provide assistance. Predictive behavior analysis for
Assisted Living might determine the probability of falls or the change of care
levels.
When movements fall outside of this pattern, an alarm can be sounded
depending on how extreme the case is.

6. Assisted Living: The Approach
Visualization Techniques
• Vector autoregression
• Markov chains
• Inter-quartile range outlier detection
• Fast fourier transform
• Random forest
Exploratory analysis – understand
the data and its nuances
Data visualization – what consists
of the “norm” within a household
and how should the said “norm”
be determined
Predictive power – use the past to
predict how the next minutes and
Page 6
• Box-and-whisker plots
• Scatter plots
• Line graphs
• Autocorrelation graphs
• Etc.
Modeling Techniques
predict how the next minutes and
hours in a house ought to be
Value extraction – what can be
learned about a household from
their activity patterns?
Can energy management be
employed?
Can generalized fingerprints for
certain groups (family size, ages,
etc) be determined?

7. Assisted Living: The Outcome
Insights obtained:
Strong patterns exist within a household, which can be broken down further into patterns by
time of day and day of the week. A household’s “fingerprint” is determined using about a
month’s worth of prior activity. This is neither predetermined nor constant – it moves with time.
Visualization of incoming activity compared to the “fingerprint” gives a first cut for any
abnormal activity.
More advanced machine learning and modeling methods delve further into the data for
patterns indiscernible by the human eye.
Page 7
Areas to be further explored:
- Tweaking the prediction methodology for outlier detection
- Use features within data, such as time of day, for increased prediction power
- Further analysis for underlying patterns in the data
patterns indiscernible by the human eye.