πŸ“ Scientific Methodology

ML Workflow & Model Architecture

Step-by-step explanation of the data pipeline, imputation strategy, ensemble learning architecture, and SHAP explainability approach used across all five Abelmoschus esculentus application domains.

Workflow Overview

Data Processing & Modeling Pipeline

πŸ“„ Literature Data
57 references Β· 231 runs
β†’
🧹 Data Curation
5 domain CSVs
β†’
πŸ”¬ MICE Imputation
Chained equations
β†’
🏷️ Encoding
One-hot / label
β†’
πŸ“Š 5-Fold CV
Stratified split
β†’
🌲 Ensemble Training
RF + GBM + ET
β†’
πŸ” SHAP Analysis
Feature attribution
Step 1

MICE Imputation β€” Handling Missing Data

Experimental datasets extracted from literature often contain missing values for unreported conditions. MICE preserves inter-feature correlations during imputation.

What is MICE?

Multivariate Imputation by Chained Equations (MICE) imputes each missing variable by fitting a predictive model conditional on all other variables, iterating until convergence. Unlike mean/median imputation, MICE produces statistically consistent estimates that reflect the joint distribution of the data.

Why Not Simple Imputation?

Features like Icorr, Rct, and Ξ”H are physically correlated (higher Icorr ↔ lower Rct). Mean imputation would break these relationships, inflating artificial correlation structures and biasing model training. MICE respects these dependencies by conditioning on all available features simultaneously.

Implementation Details

  • Python IterativeImputer (scikit-learn) with max_iter=10
  • Random Forest estimator as the internal imputation model
  • Imputed only numerical features; categorical variables encoded beforehand
  • Applied separately per domain to respect domain-specific correlations
  • Original vs. imputed comparison stored in dataset_imputation_comparison.csv

Imputation Impact

MICE imputation increased usable dataset size by allowing rows with partial measurements to be retained. The five imputed CSVs form the training set for all ML models. Domain-specific missing rates ranged from <5% (corrosion) to ~30% (composites thermomechanical properties).

Step 2

Heterogeneous Voting Ensemble Architecture

Three complementary tree-based learners are combined via soft (averaged probability) voting to reduce variance and improve generalization.

🌲 Random Forest

Bagging of decision trees with random feature subsets. Low variance, moderate bias. Robust to outliers.

n_estimators: 100
max_features: sqrt

πŸ“ˆ Gradient Boosting

Sequential boosting of weak learners with gradient descent on loss. Low bias, moderate variance. Best single-model for adsorption.

n_estimators: 100
learning_rate: 0.1

🌳 Extra Trees

Extremely Randomized Trees β€” random split thresholds reduce variance further at the cost of slight bias. Fastest to train.

n_estimators: 100
max_features: sqrt

↓ Soft Voting (averaged predictions) ↓
Heterogeneous Voting Ensemble
Final prediction = mean of 3 model outputs

Why Heterogeneous Voting?

Each base model has different inductive biases and error patterns. Combining them via averaging reduces the risk of any one model's systematic error dominating. This is especially effective for small experimental datasets where individual models can overfit.

Performance vs. Individual Models

The ensemble outperformed every individual model on all four predictable domains. The largest improvement was in adsorption: Ensemble RΒ²=0.8987 vs. GBM best=0.8841 (+1.6 pp). In electrochemistry, ensemble closed the gap between RF (0.8146) and GBM (0.8693), achieving 0.8951.

Step 3

SHAP Explainability Analysis

SHAP (SHapley Additive exPlanations) decomposes each model prediction into individual feature contributions, providing transparent, physics-consistent explanations.

What SHAP Does

For each prediction, SHAP computes the contribution of each feature by averaging over all possible feature orderings (Shapley values from cooperative game theory). The mean absolute SHAP value across all samples provides a global feature importance ranking that is additive and consistent.

Physics Validation

SHAP results were checked against known electrochemical and adsorption theory. Icorr dominance in corrosion matches the Faraday–Butler-Volmer formalism (IE = 1 βˆ’ Icorr/Icorr_blank). Ξ”H dominance in adsorption aligns with van't Hoff thermodynamics. These findings increase confidence in model validity.

Top 3 SHAP Features per Domain

DomainRankFeatureMean |SHAP|Interpretation
Corrosion Inhibition 1 ⭐Icorr_uAcm212.2479Electrochemical inhibition baseline; directly tracks IE%
2Inhibitor_Concentration_ppm7.0209Log-saturation concentration response (Langmuir-like)
3Rct_ohm_cm21.7263Charge-transfer resistance reflects adsorbed inhibitor film
Adsorption / Pollutant Removal 1 ⭐Thermodynamic_DH_kJ_mol0.9148Exothermic Ξ”H β†’ high qe; thermodynamic driving force
2Thermodynamic_DG_kJ_mol0.1734Spontaneity of adsorption (negative Ξ”G = favorable)
3Thermodynamic_DS_J_mol_K0.1550Entropy change reflects adsorbent surface randomness
Biopolymers & Composites 1 ⭐Secondary_Reinforcement_wt_percent2.3588CHB content drives tensile improvement most strongly
2Secondary_Reinforcement (type)1.9340Identity of secondary filler determines matrix compatibility
3Okra_Fiber_Content_wt_percent0.7558Fiber loading improves strength up to optimal threshold
Electrochemistry / Carbon 1 ⭐Current_Density_A_g24.6369Rate capability: higher J β†’ lower accessible capacitance
2Measurement_Configuration12.33613-electrode ≫ 2-electrode by ~47% (cell geometry effect)
3Electrolyte7.9384Ion size, viscosity, and voltage window determine charge storage
Nanoparticle Synthesis
Not predicted
1Target_Organism0.3971Biological target modulates inhibition zone size
2Nanoparticle_Size_nm0.3716Smaller NPs β†’ greater surface area β†’ higher antimicrobial activity
3Dose_ug_mL0.3276Dose-response relationship for inhibition zone
Model Evaluation

5-Fold Cross-Validation Results

All metrics are mean 5-fold CV scores. MAE and RMSE units: IE% for corrosion, log(mg/g+1) for adsorption, MPa for composites, F/g for electrochemistry, mm for nanoparticles.

DomainModelCV RΒ²CV MAECV RMSEAssessment
Corrosion InhibitionRandom Forest0.84245.36410.859Good
Gradient Boosting0.82265.30911.521Good
Extra Trees0.84084.75910.912Good
πŸ† Heterogeneous Voting0.84224.91110.865Best
Adsorption / Pollutant RemovalRandom Forest0.87740.3500.468Good
Gradient Boosting0.88410.3010.455Good
Extra Trees0.86910.3210.484Good
πŸ† Heterogeneous Voting0.89870.2830.426Best
Biopolymers & CompositesRandom Forest0.55912.8804.113Moderate
Gradient Boosting0.57352.9064.046Moderate
Extra Trees0.53602.9954.220Moderate
πŸ† Heterogeneous Voting0.57442.9154.041Best/Moderate
Electrochemistry / CarbonRandom Forest0.814612.23117.598Good
Gradient Boosting0.869310.12314.772Good
Extra Trees0.829010.75416.897Good
πŸ† Heterogeneous Voting0.89518.97413.234Best
Nanoparticle Green SynthesisRandom Forestβˆ’0.19991.1782.017Poor
Gradient Boostingβˆ’0.80281.4302.472Poor
Extra Treesβˆ’0.75251.4122.438Poor
Heterogeneous Votingβˆ’0.52681.2922.275Poor
ℹ️ Nanoparticle domain note: All models yield negative RΒ² scores, indicating predictions are worse than simply predicting the mean. Likely causes: small sample size, high experimental variability in antimicrobial assays, and heterogeneity across nanoparticle types and target organisms. This domain was excluded from the interactive predictor.

Log Transformation β€” Adsorption

The adsorption capacity target (qe) spans three orders of magnitude (3.9–1132 mg/g), making it highly right-skewed. Log-transforming the target (log(qe+1)) normalizes the distribution, improving model performance. MAE reported in log units; predictions are back-transformed via exp(Β·)βˆ’1 for display.

Composites β€” Why Lower RΒ²?

The composites dataset covers mechanically distinct systems (polyester, PP, CMC) with varying test protocols and loading directions. Pooling these into a single model introduces heterogeneity. Additionally, the small dataset (n<50) limits generalization. RΒ²=0.57 is respectable for cross-system mechanical property prediction from processing parameters alone.

References

Key Literature Sources

Corrosion Inhibition

  • Obasi et al. (2023) Environmental Science and Pollution Research β€” Okra mucilage extract as green corrosion inhibitor for mild steel in HCl (DOI: 10.1007/s11356-023-30635-0)
  • Ref 6: Corrosion current density and EIS data for okra graft polymer inhibitors in 1M HCl

Adsorption / Pollutant Removal

  • Ref 1: Okra polysaccharide for microplastic removal from natural water matrices
  • Ref 2: Okra polysaccharide for Methyl Violet 6B dye adsorption
  • Ref 3: Okra stem biochar (OSBC) for fluoride removal
  • Ref 4: Succinylated okra biomass for heavy metal ion uptake
  • Ref 5: Okra leaves for chromium(III/VI) removal
  • Ref 7: Fibrillated & EDTA-modified cellulose for Pb, Hg, Cd removal
  • Attallah et al. (2023) Reactions β€” Okra MCC/clay composite for Cu, Ni, dye

Biopolymers & Composites

  • s10751-024-02058-x: Polyester/okra fiber composites with corn husk biochar secondary reinforcement
  • s40034-019-00138-0: Polypropylene/okra composites with UV treatment effects
  • polymers-14-04884-v2: CMC/okra polysaccharide bionanocomposite films with essential oil

Electrochemistry / Carbon

  • Majhi et al. (2025) Batteries & Supercaps β€” Pore-engineered okra-derived activated carbon (OAC, SSA=2109 mΒ²/g) for supercapacitors
  • Li et al. (2019) Energy Technology β€” Natural okra shells-derived N-doped porous carbon (N-OSC, SSA=2703 mΒ²/g) for Li-S batteries
  • Molecules (2024) β€” KOH-activated okra biomass for ORR electrocatalysis
Full Citation & Dataset Info β†’