Blockchain offers cryptographically secure storage for recording transactions. However, one issue with blockchains is the problem of bad data and data reliability, where bad data refers to inaccurate, incomplete, or irrelevant data. This paper investigates how machine learning (ML) can be used to identify inaccurate sensor data added to a blockchain in Internet of Vehicles (IoV) applications. A solution for reducing the inclusion of incorrect data using a reputation-based method is proposed. We suggest that if an accurate ML model can be built for a task that can be completed using the input sensor data, it is possible to use the same model to assess the accuracy of new input data samples for which the actual task outcome is known. A road surface-type classification task is performed using Convolutional Neural Network models on the Passive Vehicular Sensors Datasets, and a pre-trained model is used in a novel solution approach involving edge servers and validators on a blockchain network. Our research shows that ML can be used to identify bad data on the blockchain and to reduce the addition of unreliable data to the blockchain in an IoV context. The proposed solution is generalizable and can be applied to any scenario where an accurate ML model can be devised for a task that can be accomplished using some blockchain input data.