The three phases where the tools live

Three phases, thirteen tasks

Everything you will do in Sheets, Excel and R belongs to one of three middle phases, and each phase breaks into named tasks. Learn the tasks, because the paper's phase questions are really task questions with the adjacent phase as the distractor.

Data understanding — look, do not change. - collect the initial data (the export, the CSV the bank sent) - describe the data (rows, columns, types — the data dictionary) - explore the data (a histogram, a first pivot, a scatter — to see, not to answer) - verify data quality (duplicates, blanks, impossible values — found, not yet fixed)

Data preparation — change the table. - select data (the rows and columns the question needs) - clean data (fix or remove what quality checks found) - construct data (a derived column: amount per unit, month, a High/Low flag) - integrate data (merge or look up across sheets: VLOOKUP of a price, merge() in R) - format data (the layout the tool needs: unpivot months-as-columns, parse text dates)

Modelling — produce the summary that answers the question. - select the technique (a pivot of group means; a fitted line) - generate the test design (how the result will be judged before it exists) - build the model - assess the model (does it fit technically — R², sensible group sizes)

Worked example — dropping the course's operations into the table

Maple & Birch's analyst, Jonah, works through the Alberta question. He writes a data dictionary for the orders export (describe) and draws a histogram of amount to see that one order, $1,320, sits far above the rest (explore). He notices the histogram is drawn in the same sheet as the data — that changes nothing: he is looking, so this is data understanding. He runs TRIM on the prov column because "ON " with a trailing space was counting as a fifth province (clean); adds amount / units as a unit-price column (construct); and uses VLOOKUP to pull each product's list price from the products sheet (integrate). Then he builds a pivot of average amount by prov (build) and checks the group counts — AB has two rows (assess).

The operation that students most often misplace is the pivot. It feels like "just summarising", so data understanding is the most-picked wrong answer. But a pivot of group means built to answer the question is the model in this course: it is the simplest comparison that could serve the decision, and it belongs to modelling.

The purpose test for borderline tasks

"Check for duplicates" can be two different tasks. If you count the duplicate rows and record that there are three, you have verified quality — data understanding. If you delete them, you have cleaned — preparation. The tool is the same; the purpose differs, and a well-written stem says which. If a stem only says "check", read it as look.

Place six

(1) Parse a text date column into real dates — format. (2) Add a column flagging orders over $500 — construct. (3) Count blank cells in amount — verify quality. (4) Drop the two test orders from the export — select or clean (both are preparation). (5) Draw a scatter of units against amount to see the shape — explore. (6) Fit a line through that scatter to predict amount — build. Then name the task for three more of your own.

CRISP-DM: this lesson is the map of data understanding → describe/explore/verify, data preparation → select/clean/construct/integrate/format, and modelling → build/assess.