Detecting Non-Linear Relationships

Not every variable behaves linearly. Many industrial processes naturally produce curved patterns that a straight-line model cannot capture.

Signs of non-linearity:

• Scatter plot shows a curve, arc, or S-shape
• Residual plot shows patterns instead of random noise
• R² increases dramatically after applying a transform
• Variable range is extremely wide (e.g. 0.1–5000)
• Process behavior changes at different operating zones

If you see these signs, try LOG, LN, SQRT, or inverse transforms.