Active feature selection cuts data burden in mental health forecasting
Active Feature Acquisition for Cost-Efficient Temporal Prediction with Reduced Participant Burden
Machine LearningArtificial Intelligence
Summary
Forecasting mental health outcomes often requires collecting many data points from people over time, which can be tiring and cause some to stop participating. The authors look at a smarter way to choose which data to collect at each time, so they can still make good predictions without asking too many questions every time. They use neural networks to decide these important data points, then turn those complex decisions into simpler trees that are easier to understand. Testing on simulated data and real alcohol use data shows they can reduce questions without losing much accuracy.
What this means in practice
- •For mental health app developers: Build apps that ask fewer daily questions while still predicting user health trends accurately.
- •For behavioral intervention teams: Design monitoring protocols that reduce participant dropout by limiting collected data without losing forecast quality.
Authors
Yunni Qu, Bing Cai Kok, Whitney Ringwald, Grant King, Aidan Wright, Kathleen Gates, Junier Oliva
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.