Source
Adapted from my own handwritten and hand-typed notes from Year 1 to Year 3 of my EEE degree
There are two primary data types:
- Structured data: highly organised, clearly defined variables, easy to mine and analyse
- Unstructured data: highly unorganised, non-obvious variables, highly context sensitive
Structured Data
| Kind | Content | Example sources |
|---|---|---|
| Numeric | continuous numeric variables (e.g. TV, radio, newspaper ad spend vs sales) | spreadsheets (Excel, CSV), SQL databases, sensors |
| Categorical | factor, level or class variables (e.g. car safety: low, med, high) | spreadsheets, SQL databases, sensors |
| Mixed | numeric and categorical together (e.g. price plus condition) | spreadsheets, SQL databases |
| Time series | numeric values with timestamps on a clear time axis | stock markets, weather over time, prices and promotions |
| Network | nodes joined by clearly defined links or edges | social networks, MRT network, financial transactions |
Unstructured Data
| Kind | Content | Example sources |
|---|---|---|
| Text | words, phrases, emoticons, hashtags | social media, WhatsApp, books, wikis |
| Image | pixels and objects | phone cameras, social media, blogs |
| Video | images, frames, objects | YouTube, video calls |
| Voice | voice signals and waves | songs, microphones, recordings, Siri |
Why It Matters
Points to note:
- structured data can go straight into tables and statistics
- unstructured data first needs feature extraction before analysis
- the amount of unstructured data is huge because billions of people use the internet, YouTube, Facebook and so on every month
The course works mainly with one mixed structured dataset, the Pokemon Dataset.