Data obsolescence detection in the light of newly acquired valid observations

The information describing the conditions of a system or a person is constantly evolving and may become obsolete and contradict other information. A database, therefore, must be consistently updated upon the acquisition of new valid observations that contradict obsolete ones contained in the database. In this paper, we propose a novel causation-based system for dealing with the information obsolescence problem when a causal Bayesian network is our representation model. Our approach is based on studying causal dependencies between the network variables to detect, in real-time, contradictions between the observations on a single subject and then identify the obsolete ones. We propose a new approximate concept, ε -Contradiction, which represents the confidence level of having a contradiction between some observations relating to a specific subject. Once identified, obsolete observations are given in an original way, in the form of an explanation AND-OR Tree. Our approach can be applied in various domains where the main issue is to detect and explain personalized situations such that the reasons and circumstances underlying unexpected outcomes. Examples include among others: detecting behaviour change by analyzing user profiles, and identifying the causes of some anomalies such as bank frauds by analyzing customer interactions. In this paper, we demonstrate the effectiveness of our approach in a real-life medical application: the elderly fall-prevention and showcase how the resulted explanation AND-OR trees can be used to give reliable recommendations to physicians and assist decision-makers. Our approach runs in a polynomial time and gives systematically and substantially good results.