New method improves budget-based feature selection for accurate predictions
Learning Source Acquisition Policies by Offline Planning
Machine Learning
Summary
Sometimes computers must decide which pieces of information to look at first because they have a limited budget or time. The authors created a method called O-MPAC that learns how to choose these pieces efficiently by studying complete examples and then making smart decisions even when only some information is available. Their approach updates its choices as more information is gathered, helping to make accurate predictions while staying within budget. Tests on several tasks show that O-MPAC outperforms other methods in balancing accuracy and cost.
What this means in practice
- •For data science teams: Improve predictive models by selecting feature groups dynamically under acquisition budget limits to maintain accuracy efficiently.
- •For network operation centers: Optimize routing decisions by sequentially querying routing sources to achieve high accuracy with constrained query budgets.
Authors
Ziqi Zhao, Run Xu, Qingjian Ni
Abstract
Predicting under an acquisition budget requires choosing feature groups whose value can depend on later queries. O-MPAC transfers finite-horizon risk-cost targets from complete training records into a shared source-action scorer. At inference time, the scorer uses partial observations and source metadata, re-scores after each query, and applies a hard cost mask. We analyze how tied teacher targets and the remaining planning horizon affect the learned decisions. Uniform supervision over tied minima preserves the target distribution under source relabeling. In a five-seed routing experiment, it achieves 0.965 accuracy under both original and context-last orders. On six real tasks, validation selects H1 without action cross-entropy in all thirty splits. O-MPAC has the highest mean budget-integrated accuracy on five tasks against source-adapted GDFS, DIME, AACO+NN and a static policy.