Identifying causal relationships is a key premise of scientific research. The growth of observational data in different disciplines along with the availability of machine learning methods offers the possibility of using an empirical approach to identifying potential causal relationships, to deepen our understandings of causal behavior and to build theories accordingly. Conventional methods of causality inference from observational data require a considerable length of time series data to capture cause-effect relationship. We find that potential causal relationships can be inferred from the composition of one step transition rates to and from an event. Also known as Markov chain, one step transition rates are a commonly available resource in different scientific disciplines. Here we introduce a simple, effective and computationally efficient method that we termed 'Causality Inference using Composition of Transitions CICT' to reveal causal structure with high accuracy. We characterize the differences in causes, effects, and random events in the composition of their inputs and outputs. To demonstrate our method, we use an administrative inpatient healthcare dataset to set up a graph network of patients transition between different diagnoses. Then we apply our method to patients transition graph, revealing deep and complex causal structure between clinical conditions. Our method shows high accuracy in predicting whether a transition in a Markov chain is causal or random, and good performance in identifying the direction of causality in bidirectional associations. Moreover, CICT brings in the new type of information that enables unsupervised clustering methods to discriminate causality from randomness. Comprehensive analysis of performance and content of computational models and comparison with medical ground truth validates our findings.
from cs.AI updates on arXiv.org http://ift.tt/2aQpZXE
via IFTTT
No comments:
Post a Comment