truncationunder

Truncationunder is a data transformation used to adjust the lower tail of a dataset relative to a specified lower bound. It truncates values that fall under the bound and has variants that differ in how the truncated values are treated. The term blends truncation with under to emphasize values beneath the threshold.

Variants include: drop, where any element x < t is removed; cap, where those elements are raised to t; and a soft variant that gradually moves values toward t without a hard cut.

Applications include data cleaning to reduce low-valued noise, sensor data preprocessing to mitigate drift, and robust statistics to stabilize estimators when small values are unreliable. It also appears in financial time series to ignore insignificant volumes.

Example: X = [0.2, 1.5, 0.5, 3.0], threshold t = 1.0. Drop yields [1.5, 3.0]; cap yields [1.0, 1.5, 1.0, 3.0].

Limitations include potential bias, distorted distributions, and altered downstream results. The choice of threshold and variant should be guided by domain knowledge and sensitivity analysis.

See also: truncation, thresholding, data cleaning, clipping.

---