Discretizing Logged Interaction Data Biases Learning for Decision-Making
This addresses a preprocessing issue for researchers and practitioners in fields like customer analytics, though it is incremental as it focuses on a specific bias in existing methods.
The paper tackles the problem of bias introduced by discretizing irregularly measured time series data for decision-making models, showing that using continuous-time models avoids this discretization bias.
Time series data that are not measured at regular intervals are commonly discretized as a preprocessing step. For example, data about customer arrival times might be simplified by summing the number of arrivals within hourly intervals, which produces a discrete-time time series that is easier to model. In this abstract, we show that discretization introduces a bias that affects models trained for decision-making. We refer to this phenomenon as discretization bias, and show that we can avoid it by using continuous-time models instead.