ICIC_Prediction:基于因果关系全局动态特性的预测方法

因果关系的预测是因果关系研究的重要内容和主要应用。现有的很多预测方法以寻找最优预测方程或最小特征变量集合为目的,以简化计算。提出一种新的可用于处理政策干预的因果关系预测方法ICIC_Prediction,不局限于利用马尔科夫毯等特征变量集合,而是从因果关系网络结构出发,利用因果关系系统及其采样数据的动态全局特性,预测目标变量在当前采样中的取值。通过在NIPS 2008"因果与预测"的评测会议上发布的四个不同类型的数据集上的对比实验,分析并展示了ICIC_Prediction方法的优势和特点。

ICIC_Prediction:A novel predictive method based on global dynamic properties of causality system

Causality prediction is an important part and a  promoing application of causality research. To simplify prediction, traditional methods usually use optimal structure equation or minimum feature set, such as Markov Blanket (MB) of the target variable, to make predictions. We present a novel method avoiding some limitations of the traditional methods by using global dynamic properties of causality analysis based on causal structure with samples to make prediction of the target variable under the effect of unknown policy manipulations performed by an external agent. Several experiments have been done to compare GC, VAR and several other popular methods with ICIC_Prediction based on four datasets published in challenge of “Causation and Prediction” in NIPS 2008 for analyzing and showing the advantages and properties of our method.