Abstract

This thesis presents a heuristic for semi-automated scientific discovery and a philosophy of science compatible with scientific knowledge discovery in databases. The heuristic may be used in conjuction with rule-based techniques to aid autonomous discovery. The heuristic function's accuracy improves with experience given a metalearning technique presented herein. The heuristics have been implemented in a program called the Scienceomatic. The model-building and metalearning capabilities of this program have been employed to search for weakly predictive patterns in seismological data