Data Visualization Workflow

DATA VISUALIZATION

WORKFLOWAn Introduction to the Art and Science of

Preparing Data Visualizations

Agenda

• Getting the data

• Exploring and cleansing the data

• Identifying key variables

• Testing for usability (normality, segmentation, etc.)

• Building the data

• Creating the model and/or visualization

Tuesday, August 25, 2015 Jeremy C. Adams 2

Getting the Data

• Where is the data? Internal to your organization?

External?

• What format is the data in? Excel file(s)? CSV? Flat files?

SQL databases? More exotic formats? (SAS, R, HANA,

Hadoop, genetic sequencing data stored in a hashed

format and encrypted to prevent tampering)

• Consider handing off data mining to a more seasoned

professional than your average software developer or

database engineer.


Getting the Data

• If you’re lucky enough to have metadata, use it! (It will

save you countless hours cleansing the data if you do.)


Exploring and Cleansing the Data

• Look at the data. You need to know how clean individual

data fields are, and which fields tend to contain data that

can be used for the visualizations you want to build.

setwd(“/Documents/Columbus Data Visualization/2015-06-26 LTC Data/raw_data")

df <- read.csv("ProviderInfo.csv")head(df) # quick view of, by default, first 10 rows of data frame



• It’s usually helpful to start by using 400 randomly chosen

rows from a given table of large size (10k+ rows).

rand400 <- df[sample(nrow(df), 400), ] # random sample of 400 rows from df



• We could combine other data with our current data set,

but for the sake of example, let’s just stick with Provider

data. We could look at number of beds by a variety of

factors.

• What if we look at Resident Occupancy rate against other

factors?

Hmmm… there should not be

10x more residents than beds!



• Let’s clean up our data (for now) by just removing data that doesn’t make sense. (In the

real world, further research would need to be done to determine how to handle this

data properly.)

Much better! And a

quick histogram reveals

what we would expect

> hist(df$RES_TO_BEDS_RATIO)


Identifying key variables

• Now comes the interesting parts – identifying how we

want to visualize our data.

• What types of ways could we visualize our data?

• Which variables contain useful data?

• There can be an art to selecting variables, and it is usually

best, and easiest, to perform direct interviews of subject

matter experts to learn about which variables may be

predictive or correlated before attempting to go directly at

the data for answers.


Testing for Usability

• Why test variables?

• Normality

• Segmentation (and segmentation biases)

• Trends

• Anomalies

• Unexpected data sorting problems

• Invalid values


Building the data

• What you build depends upon how you intend to use it.

1. Does it need to be accurate? (of course)

2. Does it need to be repeatable? (possibly)

3. Does it need to be dynamic? (optionally)

4. Does it need to be MECE? (Mutually Exclusive,

Collectively Exhaustive)

• For each of the preceding questions, expect the work to

increase… exponentially.


Creating the Model / Visualization

• Let’s review some possible examples of visualizations

based on our data

Scatter Plots



Linear Regressions

Linear Regression doesn’t even fit this data.

Perhaps having more complaints and identified

Incidents for a LTC facility helps reduce total fines?



Density Plots

Barcharts

ANOVA

Heatmaps

Geographical Heatmaps


Questions


Data Visualization Workflow

Data & Analytics