Predicting Behavior

Behavior prediction has become an important area of emphasis, with applications ranging from e-commerce, marketing analytics, and financial forecasting to smart health, security informatics, and crime prevention. However, traditional behavior modeling approaches have shortcomings: heavy reliance on objective, observed data, and a failure to consider the granular, micro-level decisions and actions that collectively drive macro-level behavior. To address these shortcomings, the authors present a behavior prediction framework that advocates the integration of objective and perceptual information and decomposes behavior into a series of closely interrelated stages to facilitate enhanced behavior prediction performance. The utility of the framework is demonstrated through a series of experiments pertaining to prediction of auction fraud, e-commerce conversions, and customer churn.