Multivariate Conditional Anomaly Detection and Its Clinical Application

Authors: Charmgil Hong, Milos Hauskrecht

AAAI 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental To validate our approach and demonstrate its effectiveness, we present experimental results on a clinical dataset obtained from Cincinnati Childrens Hospital Medical Center (Pestian et al. 2007). The dataset has 978 instances; each consists of 1,449 features (x) extracted from clinical progress notes and 45 binary class variables (y) representing the diseases diagnosed. We compared two of our chain variations chain.mod1 (Batal, Hong, and Hauskrecht 2013) and chain.mod2 (Hong and Hauskrecht 2015) with the binary relevance (BR) model (Boutell et al. 2004), which ignores the relationships between individual clinical decisions. We performed 10-fold cross validation with 3 repeats. On each round, we perturbed 15% of test data by randomly flipping 1 to 5 class variables, and see whether the methods can correctly identify the anomalies. The anomaly score is evaluated by the Mahalanobis distance on the posterior class probability P(y|x). Figure 1 shows the results in terms of the area under receiver operating characteristic (AUC).
Researcher Affiliation Academia Charmgil Hong and Milos Hauskrecht Computer Science Department University of Pittsburgh Pittsburgh, PA 15260
Pseudocode No The paper describes its methods and equations in narrative text, but no structured pseudocode or algorithm blocks are provided.
Open Source Code No The paper does not contain any statement about releasing its source code or provide any links to a code repository for the described methodology.
Open Datasets Yes To validate our approach and demonstrate its effectiveness, we present experimental results on a clinical dataset obtained from Cincinnati Childrens Hospital Medical Center (Pestian et al. 2007).
Dataset Splits Yes We performed 10-fold cross validation with 3 repeats.
Hardware Specification No The paper does not provide any specific hardware details (e.g., CPU/GPU models, memory, or computing infrastructure) used for the experiments.
Software Dependencies No The paper does not provide specific software dependencies with version numbers (e.g., Python, PyTorch, or specific libraries).
Experiment Setup Yes We compared two of our chain variations chain.mod1 (Batal, Hong, and Hauskrecht 2013) and chain.mod2 (Hong and Hauskrecht 2015) with the binary relevance (BR) model (Boutell et al. 2004), which ignores the relationships between individual clinical decisions. We performed 10-fold cross validation with 3 repeats. On each round, we perturbed 15% of test data by randomly flipping 1 to 5 class variables, and see whether the methods can correctly identify the anomalies. The anomaly score is evaluated by the Mahalanobis distance on the posterior class probability P(y|x).