Classification Accuracy of Neural Networks vs. Discriminant Analysis, Logistic Regression, and Classification and Regression Trees
Keyword(s):
Abstract. This paper compares the predictive accuracy of three commonly used parametric methods for group classification, linear discriminant analysis, quadratic discriminant analysis, and logistic regression, with two less common approaches, neural networks and classification and regression trees. The simulation study examined the impact of such factors as inequality of covariance matrices, distribution of predictors, and group size ratio (among others) on the performance of each method. Results indicate that quadratic discriminant analysis always performs as well as the other methods while neural networks behave very similarly to linear discriminant analysis and logistic regression.
1991 ◽
Vol 88
(24)
◽
pp. 11426-11430
◽
2005 ◽
Vol 15
(3)
◽
pp. 236-247
◽
1995 ◽
Vol 28
(2)
◽
pp. 287-303
◽
2021 ◽
Vol 18
(3)
◽
pp. 260-276
1996 ◽
Vol 329
(3)
◽
pp. 257-265
◽