8.3 Data Collection and Preparation
Before training, data has to be collected, cleaned, labelled and split. This prep is often most of the work.
Chapter 8: Data and AI Algorithms
Data scientists joke that the job is mostly cleaning data - and it's kind of true. Raw data is full of missing values, duplicates, typos and weird outliers.
Typical prep steps: collect (from apps, sensors, surveys, public datasets), clean (fix or remove errors), label (add the right answers), transform (scale numbers, convert text to numbers) and split into training and test sets.
Privacy matters here too: data should be collected with consent and personal details protected or removed.
Key takeaways
Collect → clean → label → transform → split
Cleaning is often the biggest chunk of work
Collect with consent and protect privacy