Emergent Future Situation Awareness: A Temporal Probabilistic Reasoning in the Absence of Domain Experts

Abstract

Dynamic Bayesian networks (DBNs) are temporal probabilistic models for reasoning over time which are rapidly gaining popularity in modern Artificial Intelligence (AI) for planning. A number of Hidden Markov Model (HMM) representations of dynamic Bayesian networks with different characteristics have been developed. However, the varieties of DBNs have obviously opened up challenging problems of how to choose the most suitable model for specific real life applications especially by non-expert practitioners. Problem of convergence over wider time steps is also challenging. Finding solutions to these challenges is difficult. In this paper, we propose a new probabilistic modeling called Emergent Future Situation Awareness (EFSA) which predicts trends over future time steps to mitigate the worries of choosing a DBN model type and avoid convergence problems when predicting over wider time steps. Its prediction strategy is based on the automatic emergence of temporal models over two dimensional (2D) time steps from historical Multivariate Time Series (MTS). Using real life publicly available MTS data on a number of comparative evaluations, our experimental results show that EFSA outperforms popular HMM and logistic regression models. This excellent performance suggests its wider application in research and industries.