Abstract: The Small Sample Size Problem （SSSP） of flights alternate is analyzed and a new prediction model of flights alternate probability distribution based on observational learning is proposed. The model uses relaxation attribute constraints to extract data subset, cubic spline interpolation method to construct the base learning, combines with the virtual data generation strategy to promote the learning agreement. And on this basis, it optimizes the confidence parameters, further improves the prediction model. Experiments of flights alternate data set show that under the large sample, the prediction model of prediction accuracy is higher than the naive Bayesian method and Bayesian network method. And on the small sample data set it analyzes the confidence under different number of alternate flights, to provide decision support for related department.

Key words: observational learning, small sample size problem, probability distribution, Bayes learning, flights alternate

摘要： 对航班备降问题的小样本特点进行了分析，提出了基于观察学习的航班备降概率分布预测模型。该模型利用松弛属性约束思想抽取数据子集，三次样条插值方法构建基学习器，并结合虚拟数据生成策略促使各基学习器达成一致。并在此基础上，对信任度参数进行优化，进一步完善了预测模型。在航班备降数据集的实验表明，在大样本下，该预测模型的预测精度高于朴素贝叶斯方法和贝叶斯网方法；在小样本数据集上分析了航班不同备降次数下的置信度，为相关部门提供决策支持。

关键词: 观察学习算法, 小样本问题, 概率分布, 贝叶斯学习, 航班备降