[R] neuralnet to discriminate a given outcome by giving cutoff outputs

[Giorgio Garziano]

I would tackle the problem in the following way:

lm.model <- lm(z~ x + y, data=m)
summary(lm.model)



Call:

lm(formula = z ~ x + y, data = m)



Residuals:

        Min          1Q      Median          3Q         Max

-0.34476713 -0.09571506 -0.01786731  0.05225554  0.51693389



Coefficients:

                Estimate   Std. Error  t value               Pr(>|t|)

(Intercept) -0.071612058  0.041196651 -1.73830              0.0876543 .

x            0.003952998  0.001336833  2.95699              0.0045417 **

y            9.968145059  0.461213516 21.61286 < 0.000000000000000222 ***

---

Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1



Residual standard error: 0.157775 on 56 degrees of freedom

Multiple R-squared:  0.905464, Adjusted R-squared:  0.9020877

F-statistic: 268.1834 on 2 and 56 DF,  p-value: < 0.00000000000000022204


coef.model <- coef(lm.model)

z.hat <- coef.model[1]+coef.model[2]*x+coef.model[3]*y

z.hat.discrete <- rep(1, length(z.hat))
z.hat.discrete[z.hat <0.4] <- 0


> all.equal(z, z.hat.discrete)

[1] TRUE


I apologise for using such naif approach in place of neural net.

--
GG

http://around-r.blogspot.it


	[[alternative HTML version deleted]]