Loading...
Loading...
Contributors:
Dataset URI:
10.5281/zenodo.23254490Overview This record contains the dataset, result tables, analysis code, and trained models for the article "Machine Learning and Deep Learning Prediction of Nanofiber Diameter in Solution Blow Spinning Using Physics-Derived Polymer Descriptors". The study predicts the fiber diameter of solution blow spun polymer fibers from eight solution, solvent, and process descriptors. It compares eight machine-learning and four deep-learning models under three validation levels: sample-level KFold and paper-level GroupKFold nested cross-validation; zero-shot prediction and post-hoc calibration on a hold-out set of 10 unseen papers; a pre-specified test on PEO fibers spun in an independent laboratory. Files SBS_dataset.csv. One row per data point. Split: CV (344 literature data points, 47 papers), Holdout (65 data points, 10 papers), or External (25 PEO conditions from the independent laboratory, 21 of them fiber-forming). Source: paper identifier, first author, year, and DOI of each literature data point. Polymer and solvent: polymer, molecular weight and its source, solvent system, Mark–Houwink constants, and solvent density. Model inputs: polymer concentration, intrinsic viscosity, c/c*, solvent surface tension, solvent boiling point, air pressure, feed rate, and working distance. Measured fiber diameter: mean, SD, and natural log. For the PEO conditions: the fiber morphology. Each PEO diameter is the mean of 100 fibers pooled across three independent runs. Data_note: flags data points that depart from the standard as-spun case, such as post-treated fibers, an applied electric field, or a diameter given as a range midpoint. SBS_results.csv. The values of the result tables in the article and Supporting Information, in long format. Columns: Analysis, Model, Item, Metric, Value, SD, Lower, Upper, and Interval. Lower and Upper give the 95% confidence interval or percentile range named in Interval. Coverage: feature summary; nested cross-validation performance; SHAP importance and feature ablation; hold-out zero-shot performance and per-paper calibration; external validation; hyperparameters; source-level clustering (ICC and variance decomposition); Level 1 sensitivity analyses; Level 2 and Level 3 calibration with five corrections; the pre-specified confirmatory test; agreement statistics. SBS_pipeline.py. The analysis pipeline, written for Google Colab and run as eight stages (S0 to S7). Data checks. Nested cross-validation of the 12 models with Optuna tuning inside every outer fold. Additional Level 1 analyses. Final models. Level 2 hold-out calibration over every anchor subset. Level 3 external validation. Tables, figures, and the numbers quoted in the text. The pipeline regenerates the results from SBS_dataset.csv. requirements.txt. Package versions used (Python 3.13). final_models_ml.pkl. The eight machine-learning models: Extra Trees, Random Forest, XGBoost, LightGBM, CatBoost, SVR, ElasticNet, and KNN. final_models_dl.pkl. The four deep-learning models: MLP, CNN1D, TabNet, and FT-Transformer. All 12 models were trained on the 344 cross-validation data points and predict ln(fiber diameter in nm) from the eight inputs. Feature scaling is part of each saved model. Both files are loaded with the load_pkl function of the pipeline, which also works without a GPU. How to run Put SBS_dataset.csv in a Google Drive folder (default MyDrive/SBS_nanofiber), open SBS_pipeline.py in Colab, and run the cells in order. The header of the file lists the run time of each stage. Licenses Data, results, and models: CC BY 4.0. Code: MIT.
Keyword(s):
MeSH Term(s):
Grants and Funding:
No Grants or Funding Data Available
Associated Publication(s):
No Associated Publication Data Available
Supplementary Data:
No Supplementary Data Available
Related Dataset(s):
No Related Dataset Data Available
The National Library of Medicine (NLM) provides information about biomedical datasets. Inclusion on an NLM Web page does not imply endorsement of, or agreement with, the contents by NLM or the National Institutes of Health. Learn more: NLM Web Policies (nih.gov)