Active feature acquisition via explainability-driven ranking

In many practical applications, including medicine, acquiring all relevant data for machine learning models is often infeasible due to constraints on time, cost, and resources. This makes it important to selectively acquire only the most informative features, yet traditional static feature selection methods fall short in scenarios where feature importance varies across instances. Here, we propose an active feature acquisition (AFA) framework, which dynamically selects features based on their importance to each individual case. Our method leverages local explanation techniques to generate instance-specific feature importance rankings. We then reframe the AFA problem as a feature prediction task, introducing a policy network grounded in a decision transformer architecture. This policy network is trained to select the next most informative feature by learning from the feature importance rankings. As a result, features are acquired sequentially, ordered by their predictive significance, leading to more efficient feature selection and acquisition. Extensive experiments on multiple datasets demonstrate that our approach outperforms current state-of-the-art AFA methods in both predictive accuracy and feature acquisition efficiency. These findings highlight the promise of an explainability-driven AFA strategy in scenarios where the cost of feature acquisition is a key concern.

In areas like medicine, gathering all the data needed for computer models can be difficult because it often takes too much time, money, or resources. This means it is important to focus on collecting only the most useful information for each situation. However, traditional methods do not adjust well when different people need different types of data. To solve this, we developed a new approach that chooses which data to collect based on what’s most important for each individual. Our method first determines which information will be most useful, then collects it step by step, prioritizing the most important pieces. When tested on different examples, our approach outperformed existing methods by making more accurate predictions while using fewer resources. This shows that focusing on the right information for each case can make data collection smarter and more efficient.