Preprocessing transforms raw data into a form suitable for analysis or modeling. Begin with the business question, then prepare data that represents the relevant slice of reality and supports reliable analysis. section 2.2 section 3.2
Before treating a dataset as ready, check whether it is:
- Relevant to the specific business question rather than merely available. section 2.2
- Accurate, complete, timely, unique, consistent, and internally coherent. section 2.2
- Structured so that missing values are handled deliberately, not automatically. section 3.2
- Prepared with appropriate data types and any variables needed for the analysis. section 3.2
A practical workflow combines relevant tables, cleans the resulting dataset, and creates analysis-ready variables. For a business dataset, this can mean joining orders, customers, products, payments, reviews, sellers, and category translations before examining sales, revenue, delivery timing, or review scores. section 3.2
A delivery-time analysis
To study delivery performance, merge the relevant order, customer, item, product, payment, review, and seller records. Handle missing values according to context, convert timestamps and numeric fields, and create a delivery-time difference between actual and estimated delivery. Then inspect summary statistics and visualizations before drawing conclusions. section 3.2 section 3.0
Cleaning is not a label to accept on trust. Inspect the original files, verify claims about quality, and confirm that the prepared structure and values support the question being asked. section 2.2
