Today, Machine Learning (ML) models are increasingly being applied in decision-making systems. The application areas of ML now range from autonomous driving, social, economical to even law. Hence, there exists an urgent need to ensure the quality of such ML models. Consequently, researchers have started to develop methods checking various sorts of requirements. However, most of the existing validation techniques either focus on a particular type of ML model (e.g. Deep Neural Network) or validating a specific property (e.g. fairness, robustness). A unified method for checking any ML model for a user specified property is largely lacking. In this work, we propose to develop a property-based testing framework for machine learning. Our approach works by generating test inputs on a white-box model via a well-known verification technique, and this white-box model is automatically inferred from the black-box model under test. On the white-box model, the space of test inputs are systematically explored by a directed computation of test cases. We term our approach as verification-based testing. So far, we have applied our technique to check the monotonicity property of ML models and find it to be more effective and efficient compared to an existing property-based testing technique.