Patient-specific inference and situation-dependent classification using Context-Sensitive Networks

Representations and inferences that capture a formal notion of "context" are needed to effectively support various analytic and learning tasks in many biomedical applications. In this paper, we formulate patient-specific inference and situation-dependent classification as context-aware reasoning tasks that can be effectively supported in probabilistic graphical networks. We introduce a new probabilistic graphical framework of Context Sensitive Networks (CSNs) to efficiently represent and reason with context-sensitive knowledge. We illustrate how different types of inference in these networks can be handled in a context-dependent manner. We also demonstrate some promising evaluation results based on a set of real-life risk prediction and model classification problems in coronary heart disease.