Data Cleaning & Regression Pipeline

Dataset

Cleaning Pipeline

Outlier fence k × IQR 3.0

Model Fit — z = a·x + b·y + c

Active rows

0 / 0

R² (fit)

—

Duplicates hidden

Outliers hidden

Legend

Clean row (complete) Missing value, imputed Duplicate row, kept Excluded from model

How it works

Each point is a synthetic data row with three numeric columns: predictors x and y, and a target z generated as z = a₀x + b₀y + c₀ + noise. Column y is sometimes missing, some rows are exact duplicates, and a few rows are seeded far outside the main cluster as outliers — exactly the mess real datasets arrive in.

Mean imputation:  y_missing ← mean(y_known)
Tukey IQR fence:  outlier if v < Q1 − k·IQR  or  v > Q3 + k·IQR
                  IQR = Q3 − Q1  (computed per column)
Least squares:    minimize Σ(z_i − a·x_i − b·y_i − c)²
Fit quality:      R² = 1 − SS_res / SS_tot

Impute Missing — rows with an unknown y are excluded from the model until you turn this on, at which point y is filled with the mean of the known values and the point becomes active (shown in amber).
Drop Duplicate Rows — hides exact-copy rows (pink) so they can't silently double-count in the fit.
Outlier fence — the pipeline recomputes Q1/Q3/IQR on every currently-active column after the missing/duplicate filters run, then hides any row that falls outside k·IQR of either fence. Lower k = stricter cleaning.
The translucent surface is the live least-squares regression plane refit to whichever rows survive the pipeline — watch it snap toward (or away from) the true relationship as you clean the data.

Real-world relevance: this is the same sequence — handle missingness, dedupe, filter outliers, then fit — that every tabular data-analysis project (pandas + scikit-learn, a BI pipeline, a spreadsheet macro) runs before any model or chart can be trusted.

Drag to orbit · scroll to zoom
Height = target z · depth = predictor y

⚠ Couldn't load the 3D engineThree.js failed to load from the CDN. Check your connection and reload.