2.76.5.1 Tutorial for Data Transformation

Notes for Starting

• Download the sample project file from here and open it in Origin.

• Start this tutorial with the app Statistical Process Control installed. If you have not installed the app, see Apps for Origin to find and install the app.

Topics for Further Reading: Introduction:Data Transformation Algorithms:Data Transformation Specifying Input for SPC Tools

User Story

A Site Reliability Engineer (SRE) is analyzing cloud infrastructure performance. He wants raw, heavily right-skewed latency metrics (API_Latency) transformed into a Gaussian distribution. So that our statistical quality control models and automated alert thresholds can reliably detect true latency spikes without producing high false-positive rates caused by skewed raw data

Steps in Origin

1. Open the sample project file in Origin, go to Folder 3. Nonnormal using the Project Explorer. Activate the workbook API Response Time
2. Highlight column B in worksheet. Click the Statistical Process Control icon in the Apps Gallery window.
3. Choose Data Transformation tab, all the data are larger than 0. From the decision tree, we should use Box-Cox method. Click Box-Cox Transform icon to open the dialog
4. In the Input tab of the opened dialog, column B will be selected automatically as Measurement Data. Set the Subgroup Size to be 1

   

5. Accept all other default settings and click OK button

Interpreting the Results

1. Double click on the embedded graph on report sheet
   • \(\lambda=0\) lies well within the 95% confidence interval [-0.549, 0.25]. It means a transformation is appropriate.
   • From the histogram and probability plot of transformed data, we can see the data are approximately normal after the transformation.
   • The transformed data follows a normal distribution (\(p = 0.894 > 0.05\)), making it suitable for parametric statistical analysis (e.g., ANOVA, hypothesis testing, or capability analysis).