A new causal discovery heuristic
ANNALS OF MATHEMATICS AND ARTIFICIAL INTELLIGENCE, vol.82, no.4, pp.245-259, 2018 (SCI-Expanded, Scopus)
- Publication Type: Article / Article
- Volume: 82 Issue: 4
- Publication Date: 2018
- Doi Number: 10.1007/s10472-018-9575-0
- Journal Name: ANNALS OF MATHEMATICS AND ARTIFICIAL INTELLIGENCE
- Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus
- Page Numbers: pp.245-259
- Hacettepe University Affiliated: Yes
Abstract
Probabilistic methods for causal discovery are based on the detection of patterns of correlation between variables. They are based on statistical theory and have revolutionised the study of causality. However, when correlation itself is unreliable, so are probabilistic methods: unusual data can lead to spurious causal links, while nonmonotonic functional relationships between variables can prevent the detection of causal links. We describe a new heuristic method for inferring causality between two continuous variables, based on randomness and unimodality tests and making few assumptions about the data. We evaluate the method against probabilistic and additive noise algorithms on real and artificial datasets, and show that it performs competitively.