The Machine Learning Process

David Chappell
PRINCIPAL, CHAPPELL & ASSOCIATES

@DChappellAssoc www.davidchappell.com
 The
Machine Learning Iterative
Process Is: - In both big and small ways

Challenging
- It’s rarely easy

Often rewarding
- But not always
The First Problem: Asking the Right Question

Choosing what Ask yourself: Do you Ask yourself: Do you

question to ask is the have the right data to know how you’ll
most important part of answer this question? measure success?
the process
 The Machine Learning Process
Choose Iterate until data Iterate to find the
data is ready best model

Raw
Data Apply Deploy
Apply pre- learning Candidate Chosen
processing Prepared chosen
Data algorithm Model model Model
to data to data
Raw
Data

Data Machine
Preprocessi Applications
Preprocessing ngLearning
Modules
Modules Algorithms
Repeating the Machine Learning Process

Re-create model
regularly

Apply Deploy
Apply pre- learning Chosen
Raw Prepared Candidate chosen
processing algorithm Model
Data Data Model model
to data to data
Scenario: Detecting Credit Card Fraud

Credit Card Customers

Payment
Application

Historical Machine
Model
Transaction Data Learning
 Scenario: Predicting Customer Churn
Customers Call Center Staff

Call Center
Detailed Aggregated CRM Data
Call Data Call Data Data for ML Application

Aggregation ML Prep
Application Application Machine
Model
Learning
Hadoop, Spark, etc.
Scenario: Predicting Equipment Failure

Devices Business Users

Streaming
Data
Real-time Data Notification
Processing Application

Historical Machine
Model
Data Learning
 Ask the right question
Choose the right data
- And get that data into good shape

The main points Iterate until you have a model that makes
good predictions
- And periodically rebuild the model

Deploy the model