๐Ÿ“– About & Citation

About the OkraML Project

Dataset overview, study background, citation guidelines, and acknowledgements for the machine learning study on Abelmoschus esculentus-derived materials.

Project Overview

OkraML: A Multi-Domain ML Study on Okra-Derived Materials

Study Motivation

Abelmoschus esculentus (okra) is an underutilized tropical crop whose various parts โ€” pods, shells, stems, leaves, seeds, and polysaccharide mucilage โ€” exhibit remarkable functional properties across diverse engineering applications. Despite numerous independent experimental studies reporting high-performance corrosion inhibition, pollutant adsorption, reinforced composites, and supercapacitor carbons, no unified predictive framework had connected these datasets.

This study addresses that gap by curating 231 experimental data points from 57 peer-reviewed references across five application domains, applying MICE imputation to handle missing values, and training a Heterogeneous Voting Ensemble (Random Forest + Gradient Boosting + Extra Trees) on each domain. SHAP analysis reveals that electrochemical (Icorr), thermodynamic (ฮ”H, ฮ”G), mechanical (secondary reinforcement), and operational (current density, measurement configuration) features are the dominant drivers in their respective domains.

Dataset Statistics

  • ๐Ÿ“Š Total experimental runs: 231
  • ๐Ÿ“š Literature references: 57
  • ๐Ÿ—‚๏ธ Application domains: 5
  • ๐Ÿ”ฌ Corrosion runs: ~67 data points
  • ๐Ÿ’ง Adsorption runs: ~41 data points
  • ๐Ÿงต Composites runs: ~44 data points
  • โšก Electrochemistry runs: ~34 data points
  • ๐Ÿ”ด Nanoparticle runs: ~45 data points

Plant Material Coverage

  • ๐ŸŒฟ Mucilage extract (aqueous polysaccharide)
  • ๐ŸŒฟ Stem biochar (OSBC)
  • ๐ŸŒฟ Pod-derived activated carbon
  • ๐ŸŒฟ Shell-derived N-doped porous carbon
  • ๐ŸŒฟ Fibrillated cellulose & EDTA-modified FC
  • ๐ŸŒฟ Succinylated okra biomass
  • ๐ŸŒฟ Okra fiber in polyester, PP, CMC matrices
  • ๐ŸŒฟ Polysaccharide gums and blends
  • ๐ŸŒฟ Green-synthesized nanoparticles (ZnO, Ag, others)
Key Findings

Summary of ML Results

โš—๏ธ
Corrosion

Rยฒ = 0.8422

Ensemble significantly outperforms baseline. Icorr (electrochemical current density) is the mechanistically dominant predictor โ€” consistent with Butler-Volmer theory. Okra mucilage achieves up to 97% IE at 25,000 ppm.

๐Ÿ’ง
Adsorption

Rยฒ = 0.8987

Highest-performing domain. Thermodynamic features (ฮ”H, ฮ”G, ฮ”S) dominate, consistent with van't Hoff adsorption theory. Wide qe range (3.9โ€“1132 mg/g) captured via log transformation. Okra polysaccharides show exceptional dye adsorption capacity.

๐Ÿงต
Composites

Rยฒ = 0.5744

Moderate performance reflects heterogeneous dataset spanning three matrix types. Secondary reinforcement (CHB) is the strongest tensile predictor. UV treatment of PP/okra composites provides substantial (+11 MPa) strength improvement.

โšก
Electrochemistry

Rยฒ = 0.8951

Excellent predictive performance. Current density and measurement configuration are overwhelmingly dominant SHAP features. OAC (SSA=2109 mยฒ/g from okra pods) achieves 318 F/g at 1 A/g โ€” competitive with commercial activated carbons.

๐Ÿ”ด
Nanoparticles

Rยฒ = โˆ’0.5268

All models fail to predict inhibition zone diameter reliably. High inter-study variability in antimicrobial protocols, nanoparticle synthesis conditions, and target organism differences make cross-study modeling infeasible with current dataset size.

Citation

How to Cite This Work

If this tool or the underlying ML study contributed to your research, please cite the associated manuscript.

APA Format

[Authors]. (2025). Machine learning prediction of multi-functional properties of Abelmoschus esculentus (okra)-derived materials: Corrosion inhibition, pollutant adsorption, biopolymer composites, and electrochemical carbon. [Journal Name]. https://doi.org/[DOI]

BibTeX

@article{okraml2025, title = {Machine learning prediction of multi-functional properties of Abelmoschus esculentus (okra)-derived materials}, author = {[Authors]}, journal = {[Journal Name]}, year = {2025}, doi = {[DOI]}, note = {Heterogeneous Voting Ensemble; MICE imputation; 5 application domains; 231 experimental runs; 57 references} }

Dataset Citation

Dataset: Okra Materials MICE-Imputed Experimental Dataset. Curated from 57 peer-reviewed literature sources covering corrosion inhibition, adsorption/pollutant removal, biopolymer composites, electrochemistry/carbon, and nanoparticle green synthesis. Total: 231 experimental data points across 5 CSV files. Associated with: [Authors], [Year], [Journal], [DOI]
Limitations & Disclaimer

Important Caveats

โš ๏ธ Prediction Uncertainty

All model predictions come with confidence intervals based on cross-validation error. The composite model (Rยฒ=0.57) carries the highest uncertainty (~ยฑ28% of predicted value). Predictions should not replace experimental determination for critical applications.

๐Ÿ“ Training Domain Boundaries

The predictor tool is only reliable within the experimental ranges covered by the training data. Extrapolation beyond these ranges (e.g., Icorr > 530 ยตA/cmยฒ, SSA > 2,800 mยฒ/g, qe > 1,132 mg/g) will produce unreliable results that violate the model's learned patterns.

๐Ÿ”ฌ Local Prediction Engine

This web tool implements a simplified regression approximation of the ensemble model, calibrated to reproduce the correct range and trends. It is not the exact scikit-learn ensemble used in the publication. For precise reproduction of paper results, use the Python script okra_ml_complete.py with the MICE-imputed CSV datasets.

๐Ÿ“š Literature-Based Dataset

The training data was compiled from diverse laboratory protocols, instruments, and reporting conventions across 57 papers. Some inter-study variability is intrinsic and cannot be fully captured. Domain-specific differences in temperature control, purity standards, and analytical methods contribute to irreducible prediction error.

โš ๏ธ Disclaimer: This tool provides research-grade predictions based on literature data collected from Abelmoschus esculentus studies. The predictions are generated by a statistical ML model and reflect the data distribution of published experiments. Results are not guaranteed to be accurate for all conditions. Experimental validation is strongly recommended before industrial, clinical, or environmental application of any okra-derived material system.
Software & Data

Tools & Packages Used

Python Environment

  • scikit-learn โ€” RF, GBM, ET, VotingRegressor, IterativeImputer
  • shap โ€” TreeExplainer for SHAP value computation
  • pandas / numpy โ€” data handling and preprocessing
  • matplotlib / seaborn โ€” visualization
  • Full pipeline: okra_ml_complete.py

Website Stack

  • Vanilla HTML5 / CSS3 / JavaScript (ES6+)
  • Google Fonts: Inter + Playfair Display
  • No external JavaScript libraries (zero dependencies)
  • Google Apps Script backend option available (apps_script_backend.gs)
  • Fully offline-capable (all predictions run client-side)

Output Files Available

  • ML_Results_Summary.csv โ€” all CV metrics
  • scratch_shap_importance.csv โ€” SHAP rankings
  • MICE_Imputed_CSV_Datasets/ โ€” 5 domain CSVs
  • Figures/ โ€” workflow, parity, SHAP plots
  • Okra_Materials_MICE_Imputed_Dataset.xlsx โ€” combined Excel