Simulation Research on Endogenous Biases in Data Stream Driven Decision Making Under Concept Drift
Abstract
There is a great deal of interest now in decision-making applications that leverage data stream mining in the context of environmental dynamism (concept drift). There has been an abundance of experimental research on adaptive data stream mining algorithms, but one challenge that does not seem to have been well-researched is the possibility that error or bias in a data stream may be endogenously generated in data stream driven decision making over time. Yet from the organizational literature on learning and adaptation, we know certain systematic blind spots can theoretically develop that would not occur if all data were analyzed simultaneously (as in hindsight). We integrate the data stream mining and organizational literatures to articulate a model within which we can assess and describe the types of errors that may occur. We propose and specify simulation-based experiments that will enable us to understand the vulnerabilities and compare solutions.
Simulation Research on Endogenous Biases in Data Stream Driven Decision Making Under Concept Drift
There is a great deal of interest now in decision-making applications that leverage data stream mining in the context of environmental dynamism (concept drift). There has been an abundance of experimental research on adaptive data stream mining algorithms, but one challenge that does not seem to have been well-researched is the possibility that error or bias in a data stream may be endogenously generated in data stream driven decision making over time. Yet from the organizational literature on learning and adaptation, we know certain systematic blind spots can theoretically develop that would not occur if all data were analyzed simultaneously (as in hindsight). We integrate the data stream mining and organizational literatures to articulate a model within which we can assess and describe the types of errors that may occur. We propose and specify simulation-based experiments that will enable us to understand the vulnerabilities and compare solutions.