cancel
Showing results for 
Search instead for 
Did you mean: 


problem type

Siya_Tshaya
NiCd Battery

 

Good day I would like to be corrected with regards to the below issue which would be a good definition of defining a problem type out of the below scenario.

Predicting whether or not a patient is likely to be readmitted on their initial discharge date is an example of the following problem type.

1.Regression

2.Text Processing

3.Time Series

4.Classification

5.Anomaly Detection

 

I would take it that it's a Regression problem type.
Could you please assist.

 

@dru 

 

Labels (1)
0 Kudos
6 Replies
JamesLee
DataRobot Employee
DataRobot Employee

@Siya_Tshaya if we are predicting a “whether or not”  outcome, that seems like we have 2 distinct outcomes and want to predict which outcome is more likely — a patient being readmitted, or not readmitted. Does that still seem like a regression target problem?

Linda
Community Team
Community Team

@Siya_Tshaya if you're looking for some reading material to help you figure this out, you can check out these:
What is Regression? (from DataRobot.com)

Confusion Matrix (for multiclass models) (from the DataRobot product docs) 

- linda

IraWatt
Micro Servo

Hi @Siya_Tshaya,

Though predicting likelihood is a percentage, your target is readmittance which is a category (was readmitted or was not readmitted) therefore it is classification.

  

Siya_Tshaya
NiCd Battery

@IraWatt I agree with you @IraWatt  but would we not also need to  include the Time series problem type since we have to consider the date(Time) the patient might be readmitted or not readmitted ?

0 Kudos
IraWatt
Micro Servo

@Siya_Tshaya the problem is if they are readmitted or not, The date they are readmitted isn't what we are looking at. Each of your entrees are independent of each other making this not a time series problem I believe. 

JamesLee
DataRobot Employee
DataRobot Employee

@Siya_Tshaya, @IraWatt is spot on -- each entry in the dataset refers to a specific example of a patient being discharged from the hospital, and then either being readmitted or not readmitted within a certain time frame. 

 

So yes, we need time/date information to help us define the target labels, but each row is independent of each other.

 

It's not a time series problem, where the rows refer to data captured for the same thing/person/entity at different points in time.