Transcript Data Exploration: Visualization

Data Driven Decision Making
Presented by
Benjamin Larson
Informatics Specialist Atlantic Health System
Founder: Analytics4all.org
Turning Data Into Decisions: Analytics
• Analytics: Discovering and communicating meaningful patterns in data
• Tools include visualization, dashboarding, data mining (machine
learning)
• The stated goal is often to develop a predictive model
• “Essentially, all models are wrong, but some are useful,” George Box
HTM and Analytics
• Many HTM professionals live in the world of descriptive analytics
▫ Static reporting, pivot tables
• The next logical step for many is the use of visualizations or even
dashboards.
• The eventual goal should be to move towards predictive or even
prescriptive analytics
▫ Machine Learning
It All Begins with Data
• Where to find data?
▫
▫
▫
▫
Your CMMS (Computerized Medical Maintenance Software)
Purchasing department
PACs administrator
3rd party parts suppliers
• Clean your data. Remove duplicates, fix errors, get data into a workable
format.
Data Exploration: Visualization
Data Visualization allows you to quickly explore your data
Simple color bars and graphs let you see areas of interest
Used extensively in descriptive analytics to allow end users
to quickly absorb vast amounts of information
Basic visualization is easy to learn and can be accomplished
with software you already have (Excel)
More advanced tools (Qlik and Tablaeu) provide more
dynamic capabilities while still remaining affordable
Data visualization can be used to create dynamic
dashboards which give you an instant understanding of
your operation.
Data Visualization: Excel
Visualization(Tablaeu): PM Work Load Balancing
Dashboarding (Qlik)
Machine Learning (Data Mining)
• Predictive Analytics
• The purpose of a predictive model is to provide you with the best guess
for an outcome
• Two main types of Machine Learning:
▫ Supervised: Data set includes a known outcomes (Decision Trees, Neural
Nets, Regression)
▫ Unsupervised: No known outcomes, so we look for patterns (Clustering,
Association)
Supervised Machine Learning
• Used to create predictive models: predicting the value of a home
▫ Machine Learning algorithms can easily work with 100’s of variables to
return results much more accurately than humans
• There are plenty of methods, each with their own pros and cons
• The only way to find the best method is through trial and error
“Who you gonna call”? – Deciding who to send out on a
call
K-Nearest Neighbor
How the Model Works
Using the data given, the algorithm creates a model
When a new call comes in, you can feed that information
into the model
Example: A call comes in for an Innova Cath Lab where the
table won’t lock (a Service Type 2). The available tier one
tech is Wilie Coyote.
If I feed this information into the model, I will get a result
like this: [0.25, 0.75]
This tells me there is a 75% chance of Tier 2 needing to be
called in.
At 75%, do I decide it just makes sense to send Tier 2 in the
first place.
Clustering – Unsupervised Machine Learning
• Clustering is a form of machine learning that
works by grouping like items
• Used widely by dating websites and
recommender algorithms
• Next I will show how I used it to review
ultrasound repair data
• For this, I used Python (Free) and Excel (which
most people already have access to)
Clustering – Ultrasound Repair Data
High Work Order / No Problem Found Cluster
This clustering high repair numbers on IE33’s was uncharacteristic.
How was it done?
I extracted 2 years worth of Ultrasound repair work orders
from my CMMS
I focused on easily quantifiable data – number of work
orders, average time spent per repair, parts costs, travel
time
I ran the data through the short program I wrote to my left.
After the program was complete, I examined the data to
see what information I could extract from it
I have a complete walk through of this on my website
https://analytics4all.org
Go to Python > Kmeans Cluster Part 1 and 2
Association – Market Basket Analysis
• Used by retailers to determine if an item is being purchased based on
other items purchased
• This can be used by HTM departments to make smarter buying plans
▫ Get all parts in one order
▫ Save on shipping
▫ Equipment downtime is shortened as multiple orders do not have to be
placed
• In the following example, I will be using XLMiner – an add-on for
Microsoft Excel
Example: Ventilators – Parts used in last 500 repairs
Data Set Sample:
Market Basket
Notice the strong relationship between power supplies
and batteries
Wrap-up
• Data collection and cleaning are where it all begins
▫ Make friends in other departments – ask for data to be sent to you in CSV
or Text form
• Data Visualization is your friend – time spent learning some basic
dataviz techniques will pay off in the end
• Machine Learning – data mining methods can churn up information you
might otherwise miss
▫ More complex though – results are easily misinterpreted
▫ Make friends with your BI department
Questions?