Analytics
PCA
Reduces many numeric columns to a few principal components — new columns PC1, PC2… that keep as much of the variation as possible.
When to use it
Use it to compress 20 sensor readings into 3 signals, to plot high-dimensional data, or to remove collinearity before clustering.
In a pipeline
File InputPCAK-Means Clustering
Settings
| Setting | Type | Required | Default or hint |
|---|---|---|---|
| Columns | select | Yes | Pick two or more numeric columns |
| Components | number | — | 2 |
| Scale columns first (z-scores) | checkbox | — | true |
| Column prefix | text | — | Default: PC (PC1, PC2…) |
Example
Temperature and humidity as one signal
- Columns=Temp, Humidity
- Components=1
| Sensor | Temp | Humidity |
|---|---|---|
| S1 | 20 | 40 |
| S2 | 22 | 44 |
| S3 | 24 | 47 |
| S4 | 26 | 53 |
| Sensor | Temp | Humidity | PC1 |
|---|---|---|---|
| S1 | 20 | 40 | -1.5962 |
| S2 | 22 | 44 | -0.5321 |
| S3 | 24 | 47 | 0.403 |
| S4 | 26 | 53 | 1.7253 |
Worth knowing
- The note says how much variance each component explains; add components until the total is high enough (often 80–95%).
- Columns are scaled to z-scores first by default, so a column in large units does not dominate; untick it for columns in the same unit.
- A component's sign is arbitrary; FlowDoo makes each component's largest weight positive so runs read the same.