On Paradox of Fuzzy Modeling: Supervised Learning for Rectifying Fuzzy Membership Function