Anomaly Detection - Unsupervised Learning without VIYA & Alternative Anomaly Detection Methods

[ertr]

Hello everbody,

I would like to ask that how can we make anomaly detection - unsupervised learning in SAS 9.4 without VIYA. Is there way to do it? If there is no way to do it without SAS VIYA license, what can be alternative methods?

The most important question is “how?”, how can make some beneficial anomaly analysis in SAS 9.4? Can someone show me with some samples? Is it possible?

And I want to prevent possible bias results, how can we understand whether the resutls may be BIAS or not?

Thank you,

[ballardw]

You might describe what sort of "anomaly" you are addressing.

Outliers, such as extreme numeric values may be determined by a number of approaches.

Patterns of values or unchanging values

Rate of change of values

Outside of control limits

Outside of instrument expected ranges

Lots of different anomaly types possible. Lots of tools depending on the type.

[Reeza]

Anomaly detection is outlier detection. You can use traditional statistical techniques and basic data steps. 

An analogy for using Base SAS is like using NotePad to type out a report compared to using Microsoft Word to write it. 

You can do it in NotePad, but it's a lot of extra work, you have to handle the model comparisons and develop the steps, whereas EM/Viya has prebuilt functions and a lot of helper tools. 

Is it possible, yes. 

Can I show you how? No. There's too many methods but I did help develop an entire fraud detection system in Base SAS that recovered millions of dollars so I know it can be done. 

Bias? I suspect you're asking about false positive and negative here but could be wrong. All models are biased, but we'd have to know your context, data collection methods and basically the details of your problem. 

---

@ertr wrote:

Hello everbody,

I would like to ask that how can we make anomaly detection - unsupervised learning in SAS 9.4 without VIYA. Is there way to do it? If there is no way to do it without SAS VIYA license, what can be alternative methods?

The most important question is “how?”, how can make some beneficial anomaly analysis in SAS 9.4? Can someone show me with some samples? Is it possible?

And I want to prevent possible bias results, how can we understand whether the resutls may be BIAS or not?

Thank you,

---