Active Feature Acquisition for Cost-Efficient Temporal Prediction with Reduced Participant Burden
Abstract
Accurate forecasting of pathological outcomes is a central problem in psychology. To do so, psychologists often collect intensive longitudinal data. However, in such studies, the desire to acquire a large number of variables for the sake of accurate prediction is often counteracted by the need to minimize participant burden. Acquiring more variables per occasion can yield better predictions, but having too many acquisitions increase the risk of non-response and attrition. Longitudinal Active Feature Acquisition (LAFA) is a principled approach to resolve this conundrum. Instead of requiring responses to every item at every acquisition occasion, LAFA produces a policy that seeks to optimally select dynamic subsets of items to be acquired at each timepoint while preserving our ability to forecast a specific outcome. However, existing LAFA methods are mostly based on Neural Networks (NN) that are difficult to interpret in practice. In this work, we introduce a tree distillation method for learning an interpretable policy from NN-based LAFA networks. We validated our method through both a simulation and an empirical EMA dataset on forecasting daily alcohol consumption. In both cases, we find that we can meaningfully reduce the number of items acquired at each occasion with minimal loss in accuracy. Networks (NN) that are difficult to interpret in practice. In this work, we introduce a tree distillation method for learning an interpretable policy from NN-based LAFA networks. We validated our method through both a simulation and an empirical EMA dataset on forecasting daily alcohol consumption. In both cases, we find that we can meaningfully reduce the number of items acquired at each occasion with minimal loss in accuracy.
Related
- Non-Myopic Active Feature Acquisition via Pathwise Policy Gradients
- JevForest: Path Voting for Budgeted Feature Acquisition
- Efficient Benchmarking Is Just Feature Selection and Multiple Regression
- Leakage-controlled Crowdfunding Success Prediction: Do Feature Selection and SMOTE Improve Boosting-Based Models?
- PPI++: Efficient prediction-powered inference
BibTeX
@misc{qu2026active,
title = {{Active Feature Acquisition for Cost-Efficient Temporal Prediction with Reduced Participant Burden}},
author = {Yunni Qu and Bing Cai Kok and Whitney Ringwald and Grant King and Aidan Wright and Kathleen Gates and Junier Oliva},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.07452},
eprint = {2610.07452},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.07452},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Active Feature Acquisition for Cost-Efficient Temporal Prediction with Reduced Participant Burden
AU - Yunni Qu
AU - Bing Cai Kok
AU - Whitney Ringwald
AU - Grant King
AU - Aidan Wright
AU - Kathleen Gates
AU - Junier Oliva
PY - 2026
DA - 2026/10/05
PB - arXiv
DO - 10.48550/arxiv.2610.07452
UR - https://arxiv.org/abs/2610.07452
AB - Accurate forecasting of pathological outcomes is a central problem in psychology. To do so, psychologists often collect intensive longitudinal data. However, in such studies, the desire to acquire a large number of variables for the sake of accurate prediction is often counteracted by the need to minimize participant burden. Acquiring more variables per occasion can yield better predictions, but having too many acquisitions increase the risk of non-response and attrition. Longitudinal Active Feature Acquisition (LAFA) is a principled approach to resolve this conundrum. Instead of requiring responses to every item at every acquisition occasion, LAFA produces a policy that seeks to optimally select dynamic subsets of items to be acquired at each timepoint while preserving our ability to forecast a specific outcome. However, existing LAFA methods are mostly based on Neural Networks (NN) that are difficult to interpret in practice. In this work, we introduce a tree distillation method for learning an interpretable policy from NN-based LAFA networks. We validated our method through both a simulation and an empirical EMA dataset on forecasting daily alcohol consumption. In both cases, we find that we can meaningfully reduce the number of items acquired at each occasion with minimal loss in accuracy. Networks (NN) that are difficult to interpret in practice. In this work, we introduce a tree distillation method for learning an interpretable policy from NN-based LAFA networks. We validated our method through both a simulation and an empirical EMA dataset on forecasting daily alcohol consumption. In both cases, we find that we can meaningfully reduce the number of items acquired at each occasion with minimal loss in accuracy.
ER -