Take the 2-minute tour ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

I have a decision (detection) problem trying to decide between symbols ${0,2}$. I have the two probability density functions: $$ f(z|s=0) = \begin{cases} 0.25z + 0.5, & -2\le\ z <0 \\ -0.25z + 0.5, & 0\le\ z \le\ 2 \end{cases} $$

and $$ f(z|s=2) = \begin{cases} 0.25z, & 0\le\ z <2 \\ -0.25z + 1, & 2\le\ z \le\ 4 \end{cases} $$

How can i mathematically prove that the optimal threshold value $T$ for that decision problem is equal to $1$?

share|improve this question
1  
Didier Piau has used the result that the maximum-likelihood decision rule minimizes the sum of the false-alarm and missed-detection probabilities. If you were to sketch the two densities, and remember that the maximum-likelihood decision rule is that if the observation has value $\alpha$, then the decision is in favor of whichever hypothesis has larger likelihood $f(\alpha\mid H_i)$, you will see immediately that the threshold is $T = 1$. Decide $s = 0$ or $2$ according as the observed value $\alpha$ is smaller than or greater than $1$. But I suppose that is not a mathematical proof. –  Dilip Sarwate Dec 13 '11 at 18:43
    
@DilipSarwate: thank you, that is enough for me! –  nikos Dec 13 '11 at 19:42
add comment

1 Answer

up vote 1 down vote accepted

One tries to maximize the probability $\frac12R(T)$ to guess right, where $$ R(T)=\int_{-\infty}^Tf(z\mid s=0)\,\mathrm dz+\int_T^{+\infty}f(z\mid s=2)\,\mathrm dz. $$ Thus, $$ R'(T)=f(T\mid s=0)-f(T\mid s=2). $$ The function $R'$ is piecewise affine, $R'(T)=0$ if $T\lt-2$ or $T\gt4$, $R'(T)=\frac14T+\frac12$ if $-2\lt T\lt0$, $R'(T)=-\frac12T+\frac12$ if $0\lt T\lt2$, $R'(T)=\frac14T-1$ if $2\lt T\lt4$.

In particular, $R'(T)\gt0$ if $-2\lt T\lt1$ and $R'(T)\lt0$ if $1\lt T\lt4$. One sees that $R(T)$ is maximum at $T=1$.

share|improve this answer
add comment

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.