Unstable approaches have been identified as the main factor in most aviation accidents, making the identification of precursors to achieve such event prediction critical for ensuring the safety and reliability of flights. However, data preparation before precursor identification is challenging due to high-dimensional variable-length time series in a specific flight phase. In this study, we propose a pipeline for flight data preparation that offers standardized inputs for the precursor mining phase and labeled outputs for the unstable approach identification phase. The raw inputs are processed by an automatic feature selection based on correlation analysis. Additionally, a uniform dynamic time warping method is proposed to transform inputs with variable lengths into equal lengths for modeling, addressing the challenge of input variability caused by different tasks and weather conditions. The effectiveness of the preparation method in flight data is validated using flight data collected from regional aircraft. It is also possible to be extended to other adverse events occurring in flight phases in terms of precursor identification.