基于气象因素与特征选择的进港航班延误可解释预测研究

Explainable Prediction of Inbound Flight Delays Based on Meteorological Factors and Feature Selection

航班延误预测对于提高旅客满意度和优化资源配置具有重要意义,然而,预测模型的不可见性限制了其进一步发展。为提高航班延误预测问题的准确性与可解释性,以某机场航线到港航班为例,在考虑机场不同气象因素的基础上,基于mRMR(max-Relevance and min-Redundancy)算法剔除冗余特征,筛选最优特征子集作为预测模型的输入,在比较多种机器学习算法后,选择Catboost算法,利用SHAP(Shapley Additive Explanation)归因分析方法,从局部解释和全局解释深入挖掘各因素对航班延误时间的不同影响程度,并采用偏依赖分析提取关键因素的最佳阈值。结果表明:经过特征选择后的Catboost预测模型能够更好地捕捉非线性特征,相比于未经过特征选择的模型,MAE(Mean Absolute Error)、RMSE(Root Mean Square Error)及MAPE(Mean Absolute Percentage Error)分别降低了3.84%,3.35%,4.22%,并利用DM(Diebold-Mariano)检验从统计学上检验模型的差异性;同时,航班延误时间受到多种气象特征以及前序延误等因素共同影响,其中,机场风速和降水量对延误时间有显著正向影响,而机场有效风力和能见度则对延误时间有显著负向影响。

Flight delay prediction is crucial for improving passenger satisfaction and optimizing resource allocation.However,the lack of visibility in predictive models hinders their further development.This paper aims to enhance the accuracy and interpretability of flight delay prediction.We focus on inbound flights of a specific airport route and develop a prediction model.We employ the max-Relevance and min-Redundancy(mRMR)algorithm to eliminate redundant features based on different meteorological factors at the airport.The optimal feature subset is then selected as input for the prediction model.The Catboost algorithm is chosen by comparing various machine learning algorithms.The Shapley Additive Explanation(SHAP)method is utilized for attribution analysis,which helps uncover the different influences of various factors on flight delay time through local and global explanations.By conducting a partial dependence analysis,the optimal threshold of key factors is extracted.The results demonstrate that the Catboost prediction model,after feature selection,performs better in capturing nonlinear features.Compared to a model without feature selection,the Mean Absolute Error(MAE),Root Mean Square Error(RMSE),and Mean Absolute Percentage Error(MAPE)are reduced by 3.84%,3.35%,and 4.22%,respectively.Statistical tests,such as the DM test,confirm the significance of the model's improvement.Moreover,the study reveals that flight delay time is influenced by various meteorological characteristics and previous delays.Specifically,airport wind speed and precipitation have a significant positive effect on delay time,while airport effective wind speed and visibility have a significant negative effect.