From: Contextual anomaly detection framework for big sensor data
Term | Definition |
---|---|
Spatial | The records in the dataset include features which identify locational information for the record. For example, a sensor reading may have spatial attributes for the city, province, and country the sensor is located in; it could also include finer-grained information about the sensors location within a building, such as floor, room, and building number. |
Graphs | The records are related to other records as per some graph structure. The graph structure then defines a spatial neighbourhood whereby these relationships can be considered as contextual indicators. |
Sequential | The records can be considered as a sequence within one another. That is, there is meaning in defining a set of records that are positioned one after another. For example, this is extremely prevalent in time-series data whereby the records are timestamped and can thus be positioned relative to each other based on time readings. |
Profile | The records can be clustered within profiles that may not have explicit temporal or spatial contextualities. This is common in anomaly detection systems where, for example, a company defines profiles for their users; should a new record violate the existing user profile, that record is declared anomalous. |