The polar transformation of a binary erasure channel (BEC) can be exactly approximated by other BECs. Arıkan proposed that polar codes for a BEC can be efficiently constructed by using its useful property. This study proposes a new class of arbitrary input generalized erasure channels, which can be exactly approximated the polar transformation by other same channel models, as with the BEC. One of the main results is the recursive formulas of the polar transformation of the proposed channel. In the study, we evaluate the polar transformation by using the $α$-mutual information. Particularly, when the input alphabet size is a prime power, we examines the following: (i) inequalities for the average of the $α$-mutual information of the proposed channel after the one-step polar transformation, and (ii) the exact proportion of polarizations of the $α$-mutual information of proposed channels in infinite number of polar transformations.