Summary form only given, as follows. Data Mining, or Knowledge Discovery in Databases as it is also called, is claimed as an offspring by three disciplines: databases, statistics, and the machine learning subfield of artificial intelligence. (The term originated in statistics with distinctly pejorative overtones - data mining was characterized as fossicking in data without a guiding model.) Statistics is obviously relevant because that field has always focused on construction of models from data. Databases, too, is clearly central because current applications of data mining can involve very large corpora of information that are not necessarily in flat file form. So what’s left to be claimed by artificial intelligence and, in particular, machine learning? This talk will provide a definitely non-impartial answer to this question from the standpoint of a long-time ML practitioner.

A reported weakness of C4.5 in domains with continuous attributes is addressed by modifying the formation and evaluation of tests on continuous attributes. An MDL-inspired penalty is applied to such tests, eliminating some of them from consideration and altering the relative desirability of all tests. Empirical trials show that the modifications lead to smaller decision trees with higher predictive accuracies. Results also confirm that a new version of C4.5 incorporating these changes is superior to recent approaches that use global discretization and that construct small trees with multi-interval splits.

First-order learning involves finding a clause-form definition of a relation from examples of the relation and relevant background information. In this paper, a particular first-order learning system is modified to customize it for finding definitions of functional relations. This restriction leads to faster learning times and, in some cases, to definitions that have higher predictive accuracy. Other first-order learning systems might benefit from similar specialization.

Various practical systems capable of extracting descriptive decision-making knowledge from data have been developed and evaluated. Techniques that represent knowledge about classified tasks in the form of decision trees are examined. A sample of techniques is sketched, ranging from basic methods of constructing decision trees to ways of using them noncategorically. Some characteristics that suggest whether a particular classification task is likely to be amenable or not to tree-based methods are discussed.>

Methods used by today's expert systems for handling inexact but valuable knowledge are discussed, and a new approach is developed that avoids some of the shortcomings.

A formalism is defined in which solutions to theorem-proving and similar problems take the form of a sequence of transformations of states into other states. The data used to solve a problem then consists of a set of rewriting rules which defines the allowable transformations. A method for selecting “useful” transformations, i.e. those which will most probably lead to a solution, is developed. Two problem-solving processes based on the above are defined; one, called the FORTRAN Deductive System (FDS) is shown to be more powerful than the other. A class of problems to which FDS will always find solutions is constructed. Examples of solutions found by a computer implementation of FDS are given and, in some cases, are compared with human performance. Finally, FDS is compared with some of the better-known problem-solving systems.