Reinforcement Feature Transformation for Polymer Property Performance Prediction

TL;DR

This work tackles polymer property performance prediction under data quality constraints by introducing GRFG, a framework that automatically reconstructs an explainable descriptor space through group-wise reinforcement learning and descriptor crossing. By partitioning descriptors into groups, employing three cascading agents, and utilizing two generation strategies, GRFG learns policies that maximize predictive performance while preserving interpretability. Empirical results on a 20-descriptor polymer dataset show GRFG outperforms multiple baselines and remains robust across different downstream models, with case studies illustrating the traceable, semantically meaningful descriptors produced. The approach offers a scalable, explainable pathway to improve polymer property predictions and can be extended to other material science tasks.

Abstract

Polymer property performance prediction aims to forecast specific features or attributes of polymers, which has become an efficient approach to measuring their performance. However, existing machine learning models face challenges in effectively learning polymer representations due to low-quality polymer datasets, which consequently impact their overall performance. This study focuses on improving polymer property performance prediction tasks by reconstructing an optimal and explainable descriptor representation space. Nevertheless, prior research such as feature engineering and representation learning can only partially solve this task since they are either labor-incentive or unexplainable. This raises two issues: 1) automatic transformation and 2) explainable enhancement. To tackle these issues, we propose our unique Traceable Group-wise Reinforcement Generation Perspective. Specifically, we redefine the reconstruction of the representation space into an interactive process, combining nested generation and selection. Generation creates meaningful descriptors, and selection eliminates redundancies to control descriptor sizes. Our approach employs cascading reinforcement learning with three Markov Decision Processes, automating descriptor and operation selection, and descriptor crossing. We utilize a group-wise generation strategy to explore and enhance reward signals for cascading agents. Ultimately, we conduct experiments to indicate the effectiveness of our proposed framework.

Figures (6)

• Figure 1: An illustration of automated transformation. We identify the optimal descriptor space that is explicit and interpretable and performs optimally in polymer property performance prediction by iteratively reconstructing the descriptor space.
• Figure 2: An overview of our proposed framework.
• Figure 3: The cascading agents include the descriptor group agent 1, the operation agent, and the descriptor group agent 2. They choose two candidate descriptor groups and a operation.
• Figure 4: Ablation study for GRFG. In this experiments, we validate the necessity of different components in GRFG.
• Figure 5: Comparison of the most important features in the original descriptor space, and the GRFG generated space.