Here’s a combinatorial version Semiclassical’s computational argument.
You have a box of white balls numbered $1$ through $n$. For some $r$ such that $0\le r\le m-i$ you pick $r$ of the balls and coler them red, and then you pick $i$ of the remaining white balls and color them blue; for a given value of $r$ there are $\binom{m}r\binom{m-r}i$ possible outcomes. Your expression is
$$\sum_{r=0}^{m-i}(-1)^r\binom{m}r\binom{m-r}i\;;\tag{1}$$
clearly this is the number of outcomes with an even number of red balls minus the number of outcomes with an odd number of red balls.
Every outcome has exactly $i$ blue balls. Let $\mathscr{I}$ be the family of all sets of $i$ balls from the box. For each $I\in\mathscr{I}$ and $r$ such that $0\le r\le m-i$ there are $\binom{m-i}r$ outcomes having $r$ red balls and $I$ as its set of blue balls. The contribution of these outcomes to $(1)$ is
$$\sum_{r=0}^{m-i}(-1)^r\binom{m-i}r\;,\tag{2}$$
and $|\mathscr{I}|=\binom{m}i$, so
$$\sum_{r=0}^{m-i}(-1)^r\binom{m}r\binom{m-r}i=\binom{m}i\sum_{r=0}^{m-i}(-1)^r\binom{m-i}r\;.$$
Finally, $(2)$ is the number of even-sized subsets of $[m-i]$ minus the number of odd-sized subsets of $[m-i]$, and it’s well known that this is $0$ unless $m=i$, in which case it’s $1$. Thus,
$$\sum_{r=0}^{m-i}(-1)^r\binom{m}r\binom{m-r}i=\begin{cases}
1,&\text{if }i=m\\
0,&\text{otherwise}\;.
\end{cases}$$