As a matter of fact, you cannot prove that
$$ V_n \stackrel?= V_{n-1} \int _0 ^{\pi/2} \cos^n \theta \, \mathrm d\theta.$$
The two sides of this alleged "equation" are not equal.
For example, let $n = 3.$ Then $V_{n-1} = V_2 = \pi$
and
$$
\int_0^{\pi/2} \cos^n \theta \, \mathrm d\theta
= \int_0^{\pi/2} \cos^3 \theta \, \mathrm d\theta = \frac23.
$$
Hence
$$
V_{n-1} \int _0 ^{\pi/2} \cos^n \theta \, \mathrm d\theta = \frac23\pi.
$$
But
$$
V_n = V_3 = \frac43\pi.
$$
To correct the equation, we can either integrate from $-\frac\pi2$ to $\frac\pi2$ instead of $0$ to $\frac\pi2$
(equivalent to integrating in Cartesian coordinates from $-r$ to $r$ instead of from $0$ to $r$, as suggested in a comment under
another answer)
or we can recognize the symmetry of that integral and simply double the original integral. That is,
$$
V_n = V_{n-1} \int _{-\pi/2} ^{\pi/2} \cos^n \theta \, \mathrm d\theta
= 2 V_{n-1} \int _0 ^{\pi/2} \cos^n \theta \, \mathrm d\theta.
$$
A correct derivation in the style of
the answer mentioned above
could be
\begin{align}
V_n
&= \int_{x_1^2+\cdots+x_{n-1}^2+x_n^2\leq 1}
\mathrm dx_1 \cdots \mathrm dx_{n-1}\, \mathrm dx_n \\
&= \int_{x_n^2\leq 1} \int_{x_1^2+\cdots+x_{n-1}^2\le 1-x^2}
\mathrm dx_1\cdots \mathrm dx_{n-1}\, \mathrm dx_n \\
&= \int_{-1}^1 \left(\sqrt{1-x^2}\right)^{n-1} V_{n-1}\, \mathrm dx \\
&= \int_{-\frac\pi2}^{\frac\pi2} (\cos^{n-1}\theta) V_{n-1}\cdot \cos\theta
\,\mathrm d\theta \\
&= V_{n-1} \int_{-\frac\pi2}^{\frac\pi2} \cos^n\theta \,\mathrm d\theta
\end{align}
using the notational convention that the volume of an $n$-ball of radius $r$ is $r^n V_n.$