Logistic regression versus neural networks for medical data
-
Autores: Gerardo Sanz Sáiz
, Á. Borque, Francisco Javier López Lorente , J.M. Vergara, Luis Mariano Esteban Escaño
- Localización: Ninth international conference Zaragoza-Pau on applied mathematics and statistics: Jaca (Spain). September 19-21, 2005 / coord. por Monique Madaune-Tort , 2006, ISBN 84-7733-871-X, págs. 245-252
- Idioma: inglés
- Enlaces
-
Resumen
In this work we consider the usefulness of classical models as Logistic regression models versus newer techniques as neural networks when they are applied to medical data. We present the difficulties appearing in the building of both types of models and their validation. For the comparison of models we have used two types of medical data that allow us to validate our models and reinforce the conclusions. Although the neural network can fit the data a little better than the logistic model, the former models are less robust than the latter. This fact together with a greater simplicity and interpretation of the variables in the logistic models makes these models preferred from the point of view of clinical applications. These results agree with other published in literature [1]
