The modeling and design of the dual active bridge converter modulation strategy are essential to achieve comprehensively optimal performance in practical applications. The commonly used modeling approaches include knowledge-based and emerging data-based methods. Despite significantly reducing the required domain expertise and analytical complexity, the data-based approach heavily relies on the quality of the collected data. Therefore, this research focuses on removing abnormal data points and ensuring data quality for the data-based modeling approach. In this paper, a data-based modeling method with data pruning (MIDI) is proposed, which employs one-class support vector machine for data preprocessing before training the surrogate model. This approach has been validated with a design case and 1kW prototype hardware experiments.