Implementing FOIL into WEKA: A Methodology and Analysis of the Problems and Benefits

摘要

Implementing a FOIL algorithm into WEKA (an open source, machine learning framework) is described The interest in implementing this algorithm into WEKA is (a) the reuse of an existing machine learning framework, and (b) the inherent problems in implementing a theoretically, radically different learning approach Since the current version of WEKA generates solutions with propositional rules, adding a FOIL algorithm would expose the benefits of generating rules that contain variables. First Order Inductive Logic algorithms can be more informative since information not overtly expressed in the data set may be generated. Further, relationships between attribute values are analyzed rather than generating a set of value ranges that result in positive coverage. The nature of the FOIL algorithm, haw to correctly represent the data, and the evaluation of a successful FOIL run are discussed.