Home

Abstract

Pancerz, K., Gomuła, J.: Two rule-based models of differential diagnosis using the MMPI test: rule decision trees and approximate rules. In: Book of Programme - Abstracts: 11th European Conference on Psychological Assessment (ECPA'2011), Riga, Latvia, August 31 - September 3, 2011, p. 111.


       In our research, we use an approach based on supervised inductive learning for cases (profiles of women examined using the MMPI-WISKAD test) grouped into eight classes: simulation, dissimulation, norm, neurosis, schizophrenia, organic, sociopathy, dependences. Data were selected for analysis by clinicians using the competent judge method. In case of a rule decision tree model, a decision tree has been generated using Quinlan’s C4.5 algorithm and, next, a rule set has been obtained. The decision tree has been pruned and decomposed and the obtained rule set has been optimized. The rule decision tree creates a model of differential diagnosis intelligible for diagnostician-clinician. In case of an approximate rule model, a rule set has been generated using algorithms implemented in the RSES system (e.g., LEM2) on the basis of cases classified beforehand. Next, this rule set has been optimized. The cross-validation methods (CV-10 and CV-5) have been used for testing the obtained rule sets. An average classification accuracy for each class was greater than 80%. A tabular collation of rules (their conditions) created for each class constitute a tabular model. Such a collation enables us to determine the so-called code types important in the MMPI diagnosis. Conditions of rules may include scales constituting a profile, indexes created on the basis of these scales as well as index systems (e.g., Goldberg’s, Diamond’s, Leary’s, Toulbee-Sisson’s, Pancheri’s, Pluzek’s). The presented functionality has been implemented in the Copernicus system. It is a tool created for computer-aided diagnosis of mental disorders based on personality inventories.