20. #MLSEV
Predictions examples
CLASSIFICATION
REGRESSION
TIME SERIES FORECASTING
CLUSTER ANALYSIS
ANOMALY DETECTION
ASSOCIATION DISCOVERY
Will this candidate meet the requirements?
How does this employee perform amongst peers?
How many customers will we have tomorrow?
Do these employees share skills?
Is this employee’s behaviour unusual?
Which traits do efficient teams share?
21. #MLSEV
So are we ready?
ITERATIVE
PREPARING AND
TRANSFORMING DATA
O
PERATIN
G
M
O
D
ELIN
G
22. #MLSEV
Scenario: call centerr
One call at a time. Will he churn?
•While asking for data, the
operator fills a form that gives
him the prediction.
Single prediction
What’s going to production?
23. #MLSEV
•Periodically predict a high amount of dataatch predictions)
Using accumulated sales data and exterior
factors (seasonality, holidays, weather
information, etc.) we want to estimate the
daily or weekly sales for each product.
Batch prediction
Scenario: Supermaket sales
What’s going to production?
24. #MLSEV
Scenario: Industrial chain
•Does the process fail?
•Sensors record current, voltage, speed,
and other physical properties. Predictions
will decide whether the process will have
faults right on the spot.
Local single predictions
What’s going to production?
25. #MLSEV
Scenario: Online Legal Assistance
•Should we sign a collection of documents?
•Users upload a handful of documents to a
web site to know whether they are correct
and can be signed.
Local batch predictions
What’s going to production?
29. #MLSEV
You need more than a model
Connecting inputs
Connecting outputs
Shift monitoring
Retrain conditions
Training the model
Model integration Model monitoring
Feedback loops