Discretizing Logged Interaction Data Biases Learning for Decision-Making

Abstract

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.

This is a standalone short paper describing a new type of bias that can arise when learning from time series data for sequential decision-making problems

References in corpus (1)

• Doubly Robust Policy Evaluation and Learning

Cited by in corpus (1)

• The Impact of Time Series Length and Discretization on Longitudinal Causal Estimation Methods