2. Chris Booker – Sales & Marketing Director
Business Analytics Forum
Introductions
3. Introductions
Chris Booker – Sales & Marketing Director
Aims of the forum
Networking
Knowledge Sharing
Community Building
Brief Introductions
Name
Role / Organisation
Interest
4. Simon Harrison 4th July 2019
Predictive Analytics
Watson Studio Walkthrough
5. Watson Data Studio
Overview of the Watson Data Studio environment
Predictive Analytics, Dashboarding, Machine Learning, AI – as a learning opportunity and
further development.
6. Watson Data Studio
Exploration Phase
Data Preparation
• Typically the most time consuming stage
• “Data Wrangling” – cleaning and shaping.
Handling inconsistencies and omissions
• User friendly
Data Exploration
• Initial Data visualisation to explore key patterns
• Reviewing initial gut feelings
Model Development
• Iterative process, generally using a range of
models to provide the strongest relationships
and most robust outputs
7. Watson Data Studio
Production Phase
Model Implementation
• Setting the baseline measures of the
effectiveness of the model e.g. scoring using a
random test dataset
Model Deployment
• Ensure the infrastructure can support
• Can it scale to support growth
Model Management
• Tracking performance over time
• Dealing with ‘Model Decay’
8. IBM Watson Data Studio
• Watson Data Studio is a platform to enter into far more advanced analytics
• The target user could be anyone from someone with a basic grounding in analytics
and would like to learn new skills, through to highly complex features for the most
capable data scientist
• Key features relevant to this session are:
• Import Data
• Refine / Cleanse
• Set data types and targets
• Select and configure model
• Run model and view findings
9. Step by step - Driver Analysis
Here is a short step by step walkthrough of how to use the platform to analyse
a customer loyalty dataset, and how to find patterns within the data.
Using the ‘Modeller’ asset type
10. Step by step - Driver Analysis
• Import Data
• Simple process to
browse for data, flat file
or to connect to
existing data sources
11. Step by step - Driver Analysis
• Data once loaded can be
profiled, quality checked and
fully catalogued
• There are a range of operations
available to refine, reclassify and
otherwise prepare the data for
analysis as required
12. Step by step - Driver Analysis
• The next thing we have to do is to set the data types
and targets
• Here we have set the Customer Lifetime Value as the
target, to repeat the earlier example of what drives the
lifetime value
• We can also exclude fields we don’t want to distort the
findings
• For this example, we have also used
the Select command to limit the
data to ‘Smart Electronics’
13. Step by step - Driver Analysis
• Select and configure
model
• A wide range of built in
algorithms are
selectable, just click and
drag onto the canvas
• Then a wide variety of
settings options
depending on which
option is chosen
14. Step by step - Driver Analysis
Run model and view findings:
16. Other Interesting Features – Machine Learning
• Each project can have
a range of assets
introduced
• Visual Recognition,
Natural Language
Classifier, Machine
Learning
• As well as SPSS
modeler, dashboards
• Deployable models
17. Simon Harrison
Analytics Operations Director
t: 0114 399 2820
m: 07949 763 848
e: simon.harrison@deeperthanblue.co.uk
https://www.linkedin.com/in/simon-harrison-dtb/
Thank you
www.deeperthanblue.co.uk
Editor's Notes
6 key stages
Important point is the model management, over time the effectiveness of any model will decay, e.g. The met office uses a 30 year average for climate figures to smooth out anomalies, it updates the data every 10 years, the next one will be next year covering 1991-2020
6 key stages
Important point is the model management, over time the effectiveness of any model will decay, e.g. The met office uses a 30 year average for climate figures to smooth out anomalies, it updates the data every 10 years, the next one will be next year covering 1991-2020
Here we start looking step by step at how to analyse a customer loyalty dataset
Starting with importing the data
Who has a google home or amazon echo type device
Looking at the output, we can see a high likelihood that our Gold loyalty status customers in suburban areas tend to buy smart electronics
This is really useful information for a marketer
For the open source fans, Jupyter notebooks are embedded in Watson studio, allowing the use of existing python code assets to be run within the studio environment
This opens up a vast range of capabilities, more than we can cover here
Within a project there are a great range of assets that can be called upon
And within the last couple of weeks there is now an Auto AI capability, as the system improves all of the time
If you have any further questions, please feel free to email me or follow or connect via LinkedIn
I would love to have your feedback and use it to improve the content
Thank you