Help center Get started Knowledge center Product documentation API documentation
Product
Lens Weave Maskbits
Overview
Data
Data Import Output Result
Data Preparation
Cleansing Imputation Duplicates Filter Fuzzy Filter Outlier Treatment Binning Trimming Validation Sorting
Transformation
Data Transformation Bulk Data Transformation Merge Union Fuzzy Merge Encoding Normalization Scaling Standardization Aggregate Unpivot Functional Transformation SQL Executor
Statistical Plots
Box Plot Correlation Histogram Pair Wise Line Plot Scatter Plot Bar Plot Stem Plot Step Plot Fill Between Stack Plot Descriptive Statistics
Workflow Agent
Workflow Agent
Scheduler
Scheduler
Library
My Workflows Shared Workflows Audit
Datasource
Datasource
Datasource Management
Data Model
Connectors Management
Admin
Admin
User Management Roles Management Group Management Organization Management

Overview

The Outlier Treatment node is a data preparation node in a Weave workflow that addresses extreme values in numeric columns. It supports three mutually exclusive methods, each with a distinct effect: IQR Method drops outlier rows, Z-score Normalization transforms all values to standardized scores, and Winsorization clamps outliers to a configured range without dropping rows.

Outlier Treatment nodes are added from the Data Preparation category of the node panel. Method selection is via mutually exclusive radio buttons — only one method applies per node. For different methods on different columns, chain multiple Outlier Treatment nodes.

Note: “Weave” is the current name for what earlier interface versions called “Workflow.” This reference uses “Weave.”

When to use it

  • Removing outliers from a dataset before downstream statistical analysis or modeling — IQR Method drops the rows which lie above or below the given quantiles.
  • Normalizing scale across numeric columns with different magnitudes — Z-score Normalization brings columns to a common standardized range.
  • Constraining outliers without losing the rows — Winsorization caps extreme values to a quantile boundary while keeping every row.

Treatment methods

Three mutually exclusive methods. The choice fundamentally changes what the node does to the data.

IQR MethodDrops rows whose values fall outside the configured quantile range. Row count decreases.Min Value and Max Value, both between 0 and 1 (quantiles).
Z-score NormalizationTransforms numeric values to Z-scores (centered at 0, scaled to standard deviation 1). Row count unchanged; values transformed.No additional parameters.
WinsorizationClamps values outside the configured quantile range to the boundary values. Row count unchanged; values capped.Min Value and Max Value, both between 0 and 1 (quantiles).

Quantile configuration (IQR Method and Winsorization)

Both IQR Method and Winsorization take Min Value and Max Value as quantile bounds. Values must be between 0 and 1; the field validates this and displays “Enter valid input (0-1)!” for out-of-range entries.

Min ValueLower quantile boundary. Values in the data below this quantile are treated as outliers.
Max ValueUpper quantile boundary. Values above this quantile are treated as outliers.

Key behaviors

Three methods, three distinct effects. Unlike Imputation where all four fill methods produce the same kind of effect (fill nulls), Outlier Treatment’s three methods do fundamentally different things to the data. IQR Method removes rows; Z-score Normalization transforms values; Winsorization clamps values. Method selection is the most important configuration decision because it determines what the node does.

Methods are mutually exclusive. Radio button selection — only one method applies per Outlier Treatment node. For different methods on different columns (e.g., Z-score Normalization on numeric features, Winsorization on a price column), chain multiple Outlier Treatment nodes.

Quantile values must be between 0 and 1. For IQR Method and Winsorization, “Enter valid input (0-1)!”. Quantiles express position in the distribution as a fraction (0.25 = the 25th percentile; 0.75 = the 75th percentile). Entering integer percentages (25 instead of 0.25) won’t validate.