Input Variable Selection in Real-time Knowledge Integration Applications: A Review, Analysis, and Recommendation Paper

The purpose of this paper is in two folds; the first is to undertake a thorough appraisal of the existing input variable selection (IVS) methods in the context of time-critical and resource-limited dimensionality reduction problems. The second is to demonstrate further improvements and the application of a recently proposed time-critical sensitivity analysis method called EventTracker in industry. Producing accurate knowledge about the state of a system (effect) in real-time under computational and data acquisition (cause) constraints is a major challenge. Especially if the knowledge required is critical in operations that the safety of operators and/or integrity of equipment is at stake. Understanding and interpreting, a chain of interrelated events, predicted or unpredicted, that may or may not result into a specific state of the system is the focus of this research challenge. The objective is to identify which set of input data/signal has significant impact on the set of system state information (i.e. output). Through this cause-effect analysis process, the proposed technique filters unsolicited data that may clog up communication and computational capabilities of a standard Supervisory Control and Data Acquisition System. The outcome of this research project is a series of issues adhered to and suggesting the difficulty of finding an established method suitable for time-critical variable selection applications. However, supported by a geological drilling monitoring application, the authors are able to substantiate the aptness of the EventTracker Sensitivity Analysis method in high volume and time critical dimensionality reduction. In order to prove the advantages gained in performance and computational efficiency by adopting the proposed sensitivity analysis method, a general comparison and evaluation of other established input variable selection techniques is conducted.