Feature engineering through two-level genetic algorithm
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
Deep learning models are widely used for their high predictive performance, but often lack interpretability. Traditional machine learning methods, such as logistic regression and ensemble models, offer greater interpretability but typically have lower predictive capacity. Feature engineering can enhance the performance of interpretable models by identifying features that optimize classification. However, existing feature engineering methods face limitations: (1) they usually do not apply non-linear transformations to features, ignoring the benefits of non-linear spaces; (2) they usually perform feature selection only once, failing to reduce uncertainty through repeated experiments; and (3) traditional methods like minimum redundancy maximum relevance (mRMR) require additional hyperparameters to define the number of selected features. To address these issues, this study proposed a hierarchical two-level feature engineering approach. In the first level, relevant features were identified using multiple bootstrapped training sets. For each training set, the features were expanded using seven non-linear transformation functions, and the minimum feature set maximizing ensemble model performance was selected using the Non-Dominated Sorting Genetic Algorithm II (NSGA-II). In the second level, candidate feature sets were aggregated using two strategies. We evaluated our approach on twelve datasets from various fields, achieving an average F1 score improvement of 1.5% while reducing the feature set size by 54.5%. Moreover, our approach outperformed or matched traditional filter-based methods. Our approach is available through a Python library ( feature-gen ), enabling others to benefit from this tool. This study highlights the utility of evolutionary algorithms to generate feature sets that enhance the performance of interpretable machine learning models.
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.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.001 | 0.000 |
| Research integrity | 0.000 | 0.001 |
| Insufficient payload (model declined to judge) | 0.000 | 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 it