关键词: 因果推断, 时间序列, 假设检验, 参数优化, 自适应

Abstract: The recency characteristic was exhibited by time-series data, where variable values were generally dependent on recent historical information. This characteristic was not fully considered by existing methods, which used a uniform threshold when causal relationships with different delays were inferred through hypothesis testing, making it difficult to effectively infer weaker causal relationships. To address the aforementioned issue, a method for time-series causal inference based on adaptive threshold learning was proposed: first, data characteristics were extracted, then, based on the nature of the data at different delays, a combination of thresholds used in the hypothesis testing process was automatically learned. Finally, this threshold combination was applied to the hypothesis testing processes of the PC (Peter-Clark) algorithm, PCMCI (Peter-Clark and Momentary Conditional Independence) algorithm, and VAR-LINGAM (Vector Autoregressive Linear non-Gaussian Acyclic Model) algorithm to obtain a more accurate causal relationship structure. Experimental verification was conducted on a simulated dataset: on dataset a, the adaptive PC algorithm, adaptive PCMCI algorithm, and adaptive VAR-LINGAM algorithm using the proposed method improved the F1 score by approximately 1.31, 1, and 0.03 percentage points respectively; on dataset b, they improved by approximately 0.53, 1.16, and 2.36 percentage points respectively; on dataset c, they improved by approximately 0.22, 3.56, and 0.98 percentage points respectively.

Key words: causal inference, time series, hypothesis test, parameter optimization, adaptive