Knowledge discovery in a water quality database

Abstract

We apply rule induction to mine for knowledge in a database which stores data obtained by monitoring the water quality of the rivers in Slovenia. The database contains measurement data about the physical and chemical properties of the water at different measurement sites, as well as data about the presence of living organisms. Taken together, the above data reflect the quality of the water: the physical and chemical properties influence the living organisms, which in turn give an overall picture of the water quality over a period of time. We address two problems: (a) analysis of the influence of physical and chemical indicators of water quality on the presence of selected bioindicators, and (b) classification into quality classes based on either bioindicator or physical and chemical indicator data. The learned rules are evaluated by a river biology expert, but also in terms of their performance on unseen cases.