Interpretable rule learning for dementia
What is it about?
A framework adopting a tailored neural network is proposed to learn interpretable rules for the prediction and early diagnosis of dementia based on cognitive tests. An intuitive visualization form of rules is also proposed for doctors. Results on two datasets verify that we can build accurate and interpretable rules with only cognitive tests.
Why is it important?
1) To our knowledge, this is the first study using Rule-based Representation Learner (RRL) to build interpretable and accurate rules for dementia prediction and early diagnosis. 2) The rules learned by RRL offer a better trade-off between classification performance and model interpretability than other representative machine learning models. 3) The novel visualization form we proposed makes the learned rules more intuitive and convenient to use for doctors. 4) Since they only depend on cognitive tests, the diagnostic rules obtained in our study are easy to promote, especially in low- and middle-income countries.