Given the intermittent nature of the wind, accurate wind energy forecasting is significant to the proper utilization of renewable energy sources. In recent years, data-driven models based on past observations have been widely employed in the literature. Various types of data processing methods are successfully applied to assist these models and further improve forecasting performance. Comprehensive research of their methodologies is called on for a thorough understanding of current challenges that affect model accuracy and efficiency. To address the knowledge gap, this paper presents an exhaustive review and categorization of data processing in wind energy forecasting. The utilized techniques are classified into seven categories according to the applications: decomposition, feature selection, feature extraction, denoising, residual error modeling, outlier detection, and filter-based correction. An overall analysis of their intentions, positions, characteristics, and implementation details is provided. A general evaluation is carried out from different perspectives including accuracy improvement, usage frequency, consuming time, robustness to parameters, maturity, and implementation difficulty. Among the existing data processing methods, outlier detection and filter-based correction are relatively less used. Their potential can be better explored in the future. Furthermore, some possible research directions and challenges of data processing in wind energy forecasting are provided, in order to encourage subsequent research.