Diagnosing Autism using ICD-10 criteria: A comparison of neural networks and standard multivariate procedures
- 1 April 1995
- journal article
- research article
- Published by Taylor & Francis in Child Neuropsychology
- Vol. 1 (1) , 26-37
- https://doi.org/10.1080/09297049508401340
Abstract
In a sample of 976 consecutive cases derived from the recent world-wide Field Trial of Autism and other Pervasive Developmental Disorders, we tested the accuracy of the 15 ICD-10 criteria for the diagnosis of Autism, by comparing neural network models (NN) to more conventional multivariate competitors, namely, linear and quadratic discriminant function analyses and logistic regression. NNs were less accurate than competitors, both in terms of cross-validation results as well as in levels of shrinkage from training to test conditions. The clinical research implications of these results are discussed.Keywords
This publication has 19 references indexed in Scilit:
- Hinton diagrams: Viewing connection strengths in neural networksBehavior Research Methods, Instruments & Computers, 1994
- Initializing back propagation networks with prototypesNeural Networks, 1993
- A neural network approach to the classification of autismJournal of Autism and Developmental Disorders, 1993
- Evaluating black-boxes as medical decision aids: issues arising from a study of neural networksMedical Informatics, 1990
- The recent excitement about neural networksNature, 1989
- An evaluation of the autism behavior checklistJournal of Autism and Developmental Disorders, 1988
- The logistic regression analysis of psychiatric dataJournal of Psychiatric Research, 1986
- A methodology for construction of a multivariate diagnostic instrument: An application to alcohol abuse screeningComputers and Biomedical Research, 1982
- Linear Discriminant Function analysis in Neuropsychological Research: Some Uses and abusesCortex, 1978
- Estimation of Error Rates in Discriminant AnalysisTechnometrics, 1968