Inference Engine driven by Situation-Based Correlation

Abstract

Drawing an intelligible inference is a challenging aspect of correlation studies in data mining. Regression analysis and inferences drawn based on correlation of system state play important role in decision making. However, traditional regression algorithms operate on data in an opaque manner thereby shielding end user from knowing the reasoning behind the inference drawn. Such techniques also fail to learn from repetitive historical conditions occurring in the system over a longer time-span. In this paper, we propose a novel situation- based correlation technique which can be used to not only predict system behavior but also to convey reasoning behind the prediction. “Situation” can be defined as a more inclusive version of the system state, which encompasses variables, parameters, rules, and relationships that describe the behavior of the system over the span of finite time interval. The proposed algorithm identifies similar situations in high dimensional time- series records and produces interpretable digital record of matching situations. We then deploy the proposed situation- based correlation algorithm as core of inference engine to successfully demonstrate fully functional expert system. (Abstract)