A "basic form" may seldom appear in a highly inflected languages, and languages in the world are different from each other, by which we cannot simply apply the convention of IE languages on an "exotic" language to define the "basic form" of a word class. Does each such language have a unique optimal way to define "basic form", or is there any cross-linguistical general rule, or any convention made by linguists, to define the basic form of a highly inflected language?
"Basic form" is defined by the linguists working on a language, and not by the language itself. Heuristically speaking, the tendency is to find some form that best allows all related forms to be calculated by analogy. Some traditions don't limit the analysis to specific words, for example "roots" are not words, but Arabic linguistics uses "roots" e.g. šrb. Sanskrit uses highly sophisticated root and stem abstractions (they had a few millenia to work out the system). In North Saami lexicography, for example, they use the nominative singular (N) or infinitive (V) but use special annotations to indicate change e.g. "Nom. sg. t becomes hp" (in contexts that you are expected to know).
Actual practice is very much influenced by the extent to which "general readers" might use such information, therefore for most Bantu languages, since non-linguists rarely use such materials, they can use underlying forms or other abstractions. The Logoori dictionary uses imperatives of consonant-initial verbs but infinitives of vowel-initial verbs, and since the author is deceased we can't ask why not use the imperative for every verb. For demonstratives, every actual form is listed (the author was not a linguist), whereas for nouns, just the singular is listed. The majority of highly-inflected languages are under-described, so that the idiosyncratic practices of one or two linguists dominate.