Types of datasets; spatial, temporal data

A data set is a collection of data objects (records) described by attributes. Data sets are grouped into three broad categories (Tan, Steinbach & Kumar).

1. Record data

Each object is a fixed set of attributes with no explicit relation between records.

• Transaction (market-basket) data: each record is a set of items bought, e.g. {bread, milk}.
• Data matrix: numeric records seen as points in a multi-dimensional space (rows = objects, columns = attributes).
• Document-term matrix: each document is a vector of word counts.

2. Graph-based data

• Data with relationships among objects: web pages linked by hyperlinks.
• Data with objects that are graphs: chemical molecules (atoms as nodes, bonds as edges).

3. Ordered data

• Sequential data: ordered sequences of transactions or events, e.g. customer purchase history.
• Sequence (genetic) data: DNA strings of A, C, G, T.
• Time series data: values measured at successive times.
• Spatial data: objects with a position in space.

Spatial data

Data whose attributes include location or geometric information, such as coordinates, shapes and distances. Nearby locations tend to have similar values (spatial autocorrelation). Examples: maps and satellite images, rainfall or temperature at weather stations, GPS locations of towns, land-use data in GIS.

Temporal data

Data in which the time dimension is important, so values change or are recorded with time. Observations close in time are related (temporal autocorrelation). Examples: monthly sales, daily stock prices, hourly electricity load, ECG signals, daily temperature of Kathmandu.

Spatio-temporal data has both, e.g. monthly average temperature at many places on the earth.