Internet of Things trifft auf Customer Intelligence

COPYRIGHT © 2014 BY PHIL WINTERS. ALL RIGHTS PROTECTED AND RESERVED.
The Internet of Things
Aaron Hart
Dr. Rosaria Silipo
Phil Winters

COPYRIGHT © 2014 BY PHIL WINTERS. ALL RIGHTS PROTECTED AND RESERVED. 2
Customer Strategy
Customer Perspective
Chamption

The Customer’s Perspective
Decision Cycle and Touchpoints
g
@
Ad
g
g

• Defining the Customer Perspective
• Identifying new Touchpoints
• Using new Customer Insight
• Give to Get: Information Strategies
• Social Media
• Mobile
• Big Data
• Sales Cycle Transformation
• Touchpoint Choreography
• Customer Delight
• Multi-Industry / Multi- Product processes optimization
Selected Topics
Doing Business from the Customer’s Perspective

Where you can find more:
Customer IMPACT Agenda
The Speeches
The Book
The Workshop
www.ciagenda.com
phil.winters@ciagenda.com
auch auf Deutsch!

Customer Strategy
Data Whisperer Customer Perspective
Chamption

Customer Intelligence – creating new fact-base insight from data
Data Whisperer
Needs
Behavior
Value
drivesgenerates
Dimension Needs Dimension Behavior Dimension Needs

PAW 2012: Text Mining meets Network Mining
8
Text Mining for Sentiment
Drill Down on special cases
Network Mining for Relevance
Analytics for Prediction

• Telemetry Data
• Time Series Analysis
with clustering
• Measurable / Applied to the Business
• SENSIBLE usages of Big Data
PAW 2013: Time Series meets Machine Learning

2014: It’s all Geert’s Fault: The Internet of Things
Illustration by CRISTINA BYVIK
Use Public Data Please….

Original:

Washington DC

Sensors!

KNIME and the Internet of Things

Reading Sensor data is always messy……
14 Quarters of Sensor Data….
But good packages
make it easy to do !

Reading the Sensor Data

Street Maps
Weather
Holiday Schedules
Commuters
Tourists
Topology / Elevations
Enrich

Enrich
The power of REST services and the Internet…………

Enrich

Expand

Station and Bike Facts
Over 3 years
307 Stations
2963 Bikes
19.4% Casual Bikers
5.9m Bike Moves

The 30 minute rule…
Under 30 30 and Over
Casual 63% 37%
Subscriber 98% 2%
Overall 91% 9%

Most Popular Segment:
Jefferson Memorial to the Lincoln Memorial
89% Casual Users!

The Business Challenge:

Even MORE of a Business Challenge
Any Station without bikes for 1 hour:
$XXXX Per Violation
Any Station with no free slots for 1 hour:
$XXXX Per Violation

The Business Challenge: 12th and Bell St.
-587 Bikes

The Business Challenge: 12th Street and Bell

“Helpful” Bike Enthusiasts
http://www.cabitracker.com/status.php

http://bikeportland.org/2013/03/10/behind-the-scenes-of-capital-bikeshare-84006
Capital Bikeshare Response

Station Totals per hour

Creating the Video…..
• Open Street Map (OSM)
• Image Processing
• Group Loop

Station Totals per hour (the video)

Lean Restocking Alert System
• Goal: Implement an alert signal for when shifting stock (both
adding and removing bikes) may be needed in a station.
• Target Variable:
Flag and Shifted contain the current human operated
restocking information.
• We want 1 hour warning!
Lag(flag-1)

29 Input Features!
- Weather related features
- Number of registered and casual people showing up
- Station info (name and max. number of docks)
- Calendar info (working day, holiday, date)
- Count as the number of bikes added and removed at
each hour
- Past bike ratios over time at that station
• Adjusted cumulative sum as the number of bikes available at
the station at a given hour
• Bike ratio = adjusted cumulative sum/total docks available.
Predict an Alert 1 hour in advance of a “full” or “empty” situation
Lean Restocking Alert System

Linear Correlation on the Input Data Columns

Feature Elimination Loop
Two options:
1. Use all the input features
(no thinking required, just a powerful machine)
2. Select the most useful input features via “Feature Elimination”
At each step one
input feature is
removed- i.e. the
input feature
whose removal
produces the
smallest error
increase.

“Backward Feature Elimination Filter” node

“Lean Restocking Alert System” workflow
The input feature subset with the
smallest error (81% accuracy) is
forwarded to the final model
training block:
• Hour of the day
• working day (Y/N)
• Current Bike Ratio
• Terminal (station code)
past Bike Ratios and weather info
does not seem to be relevant!

Total Bikers Predicted by Hour of days
• Registered (blue) vs. Casual (red) Bikers

Time Series Prediction
- Define Time series lags by hand or……
- Select the best values programmatically using an Optimization
Parameter Loop!

Parameter Optimization Loop
Looping over different values for
lagging and seasonality index
(Brute Force or Hillclimbing).
Collecting the RMS error at the end
of the loop.
Selecting the parameter set with
minimum RMS error.

“Bikers Prediction” workflow
Parameter
Optimization
Loop
Time Series Prediction Metanodes
Seasonality
index
Past lags RMS Error
casual 1hour 20 13.8
registered 24 hours 10 68.3

Conclusions….
- Best Input Feature Subset using the
“Feature Elimination” Metanode
- Best Parameter Set for Time Series Prediction
using the “Parameter Optimization” node
- Weather influence on bikers is overrated!
- Casual Bikers 24 hours Seasonality is also not so
relevant!

Top Net Bike Change Stations

Stations with deficits and surpluses
Bike sources
Bike Sinks
16th and Harvard
Union Station

NetChanges
Total Traffic Volume
Bike sources
Bike Sinks
16th and Harvard
Union Station
And now some analysis…

Finding route data
(Just google it)

REST Webservices with KNIME
http://maps.googleapis.com/maps/api/directions/json?
&origin=lat,long&destination=lat,long&mode=bicycling
https://developers.google.com/maps/documentation/directions/

Then Plot the Routes!
Union Station

Finding route data
(Just google it)

Convert to a Network for Network Analysis

Urban/
Students
Suburban
Tourists
Top 250 Routes
Dupont Circle
Union Station

Union Station Subnet

Dupont Circle Subnet

Possible Next Steps
• Combine Prediction with Path !!!!
• Enrich through analyzing images
+

IOT Lessons Learned
• IOT (Sensor) Data is simple… but can still be messy
– get it in, get it clean!
• Connect it!
– Common keys, Lat/long, time, etc.
– No common key? Network analysis + Imaging
• Enrich / Expand
– Transformation
– External Sources; REST calls; Palladian, etc.
• Explore
– Visualization (over time!) OSM, Imaging
– Network Analysis
– Correlation matrix!
• Prediction
– Feature Elimination for Input Feature Set + “Classic Machine Learning Techniques”
– Parameter Optimization + “Classic Time Series Techniques”

It’s all Geert’s Fault: The Internet of Things
Illustration by CRISTINA BYVIK
Use Public Data Please….

The Internet of Things:
White paper, Workflows, Data are now available
Taming the Internet of Things with KNIME:
Data Enrichment, Visualization,
Time Series Analysis, and Optimization
http://www.knime.com/white-papers#IoT

Internet of Things trifft auf Customer Intelligence

Recommended

Recommended

More Related Content

What's hot

What's hot (20)

Viewers also liked

Viewers also liked (6)

Similar to Internet of Things trifft auf Customer Intelligence

Similar to Internet of Things trifft auf Customer Intelligence (20)

More from Rising Media Ltd.

More from Rising Media Ltd. (20)

Recently uploaded

Recently uploaded (20)

Internet of Things trifft auf Customer Intelligence