A Probabilistic Framework for Validating Sensor-Based Data and Deriving Knowledge From Massive Datasets: Examples From Ongoing Research

Current and future collaborative research projects will increasingly rely on networked databases for managing massive data sets generated by sensor-based data collection. Such data sets may be used for developing agroclimatic models, decision support systems, and other knowledge-based applications. Bayesian belief networks (BBNs) are especially useful for model-based validation of sensor data as well as deriving knowledge from such data. BBNs graphically and probabilistically describe the influence of predictor variables on response variables. This presentation will demonstrate the use of a BBN software package for real-time validation of sensor-based agroclimatic data as well as testing of models using data sets stored in networked databases. Examples of each application will be drawn from an ongoing SCRI-funded research on the development of a model-based decision support system for sweetpotato production.