Card-based cryptography uses physical cards to realize cryptographic protocols, such as secure multiparty computation. At the last SecITC, card-based arithmetic operations using integer commitments were proposed with their application to statistical processing. In statistical processing, the statistical value increases as the number of data points increases, so the number of cards required for processing also increases. To address this issue, this paper introduces a new representation of commitments, named floating-point numerical commitments, for floating-point numbers. The floating-point numerical commitments allow us to represent a wide range of numbers, including negative and arbitrary-precision integers. We also propose a protocol that converts a conventional integer commitment to a floating-point numerical commitment and protocols for arithmetic operations using floating-point numerical commitments as inputs. We then discuss that our new representation and operations enhance the application of card-based cryptography to statistical processing.