Login

Publications  •  Project Statistics

Glossary  •  Schools  •  Disciplines
People Search: 
   
Title/Abstract Search: 

Dissertation Information for Michael Lewis

NAME:
- Michael Lewis
- (Alias) Charles Michael Lewis

DEGREE:
- Ph.D.

DISCIPLINE:
- Psychology

SCHOOL:
- Georgia Institute of Technology (USA) (1986)

ADVISORS:
- None

COMMITTEE MEMBERS:
- None

MPACT Status: Incomplete - Not_Inspected

Title: IDENTIFICATION OF RULE-BASED MODELS

Abstract: This thesis develops a methodology for the use of production (condition (--->) action) rules in the analysis of behavioral data. Two sets of data were analyzed and variations in methodology evaluated. INDUCE 3.0, a machine learning program, was used for rule identification. Machine learning programs of this type treat observations as logical assertions. The "rules" are identified by searching this space for the most general consistent expressions of the same (condition (--->) form).

A six stage analysis procedure was developed. In stage 1, data collection, it is necessary to provide stationary data in which conditions are completely represented for each observation. Stage 2, coding, requires the recording of continuous variables into a small number of ranges. In stage 3, identification, a machine learning program is used to identify rules. At stage 4, organization, the rules are organized into trees in accordance with overlapping coverage among their conditions. In stage 5, selection, these trees are used as a guide in selecting a representative set of rules. These rules should cover most of the observations covered by the complete rule set but with minimal overlap among conditions. In stage 6, testing, statistical significance testing techniques are used to ensure that identified consistencies represent characteristics of the data rather than opportunism of the search procedure.

The major import of this work lies in the application of logical induction to data analysis and the introduction of appropriate significance testing methods. In this thesis a randomization test and a cross validation procedure using (chi)('2) were used. An alternate form of randomization testing based on combinatorial ennumeration is described in the appendices.

Simple induction procedures such as the Aq and ID3 algorithms appear best suited for data analysis. The techniques developed appear widely applicable to analysis of data in the behavioral sciences. Rule-based analysis should be considered complementary to analyses based on the effects of variables such as Analysis of Variance of Multiple Regression Analysis. Rule identification is demonstrated to identify consistencies which cannot be detected by other methods.

MPACT Scores for Michael Lewis

A = 4
C = 15
A+C = 19
T = 4
G = 1
W = 4
TD = 4
TA = 0
calculated 2012-07-31 13:13:49

Advisors and Advisees Graph