WrapperRL: Reinforcement Learning Agent for Feature Selection in High-Dimensional Industrial Data
Bibliographic record
Abstract
Finding the set of discriminatory features in a classification task is imperative for the interpretability of the “black box” deep learning (DL) models, especially in high-stakes industrial applications such as predictive maintenance and industrial noise classification. In cases with time-series Time-Frequency (TF) domain data, the interpretability of DL models is challenged by the data’s high dimensionality and the need to maintain the characteristics of the original signal when interpreting classification results. This paper devises a three-stage process that supports the interpretability of a DL model identifying industrial noise through a forward feature selection procedure. The first stage transforms the original TF data into an image representation. The second stage proposes a 2D Convolutional Neural Network (CNN) with a self-attention mechanism (SA-CNN) that classifies the data into instances with and without industrial noise. The final stage, termed WrapperRL, utilizes a Reinforcement Learning (RL) agent, to find the set of discriminatory frequency bands contributing to classification results. SA-CNN and WrapperRL both outperform the state-of-the-art implementations, each in their own specialty. The insights provided by WrapperRL suggest the contribution of around 20% of frequency bands to the existence of industrial noise, mainly residing in the low-frequency domain. Together, both of these approaches serve as a promising starting point for enhancing the interpretability of DL models and explaining the classification results of industrial TF data.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.004 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.002 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".