Statistical Quality Control

ratheesan

Hi,

Is there  any option for workout statistical Quality Control in RM?

Thanks
Ratheesan

fischer

Hi,

yes. We have, e.g. T-Test and ANOVA operators.

Best,
Simon

ratheesan

Thanks Simon,
But I mean,control charts like Xbar chart,Rcharts...

Thanks
Ratheesan

IngoRM

Hi,

you can make the necessary calculations yourself with the operator "Generate Attributes" and use for example the series plotter for the x-bar chart. As far as I can see this should not be too difficult - or at least possible 

Cheers,
Ingo

Unicorn_rider

Hi,
Any update on control chart and visual tools to predict outlier that would be part of a package? Box plot are nice but it cannot help to predict a trend...

MartinLiebig

Hi @Unicorn_rider ,

i think the main problem is not the visualization here. For me the main problem is: How do I detect changes in process or changes over time? How do I know that my process is out of control.

One way to do this is to treat this as a unsupervised anomaly detection. You compare your current window of time to a reference window in time and calculate the confidence, that they are the same. This can be done using for example Windowing Tukey Test operators.

I am also thinking about publishing an operator doing CUSUM test ( https://en.wikipedia.org/wiki/CUSUM ) and Chow Test ( https://en.wikipedia.org/wiki/Chow_test) which are more tradtional methods to work on this. Maybe this would be helpful for you?

BR,

Martin

Telcontar120

The Statstics extension available in the marketplace also has operators for a number of common statistical tests, you might want to check that out.  As noted above, creating the types of visualizations often used in SPC is something you can do pretty easily in RapidMiner, and Generate Attributes also lets you compute your own calculations manually, but there is no extension specifically devoted to SPC methods.  Are there specific exhiibts or techniques that you have in mind that you are not able to see how to replicate them inside RapidMiner?