Estimating Conditional Transfer Entropy in Time Series Using Mutual Information and Nonlinear Prediction.

Area of Science:

• Time Series Analysis
• Statistical Modeling
• Neuroscience

Background:

• Accurate measurement of directed dependencies in time series is crucial for understanding complex systems.
• Existing methods often struggle with highly correlated data and false positive detections.

Purpose of the Study:

• To propose a novel estimator for measuring directed dependencies in time series data.
• To enhance accuracy and reduce false detections compared to existing state-of-the-art methods.

Main Methods:

• A new non-uniform embedding technique for dimensionality reduction, ranking variables by information and prediction accuracy.
• A greedy, iterative approach to select the most informative variable subsets.
• Comparison with existing methods using simulation studies at varying data lengths and dependency strengths.

Main Results:

• The proposed estimator demonstrated significantly higher accuracy than existing methods, particularly for highly correlated and coupled data.
• The estimator showed a lower false detection rate of directed dependencies caused by instantaneous couplings.
• Successful application of the estimator on real intracranial electroencephalography (EEG) data was demonstrated.

Conclusions:

• The novel estimator offers superior performance in identifying directed dependencies in time series.
• It provides a more reliable tool for analyzing complex, correlated data, including neurophysiological signals.