Predict material properties of okra-derived systems across corrosion inhibition, pollutant adsorption, biopolymer composites, and supercapacitor electrochemistry using a Heterogeneous Voting Ensemble ML model.
Each domain is modeled with a Heterogeneous Voting Ensemble (Random Forest + Gradient Boosting + Extra Trees), trained on MICE-imputed experimental data.
Predict inhibition efficiency (%) of okra extract and graft polymer inhibitors in HCl medium. Icorr dominates prediction.
Target: Inhibition_Efficiency_percent (0–97%)
Predict adsorption capacity qe (mg/g) for heavy metals, dyes and microplastics. Thermodynamic ΔH is the top SHAP driver.
Target: Adsorption_Capacity_qe_mg_g (log scale)
Predict tensile strength (MPa) of okra-fiber reinforced polymers. Secondary reinforcement content drives performance.
Target: Tensile_Strength_MPa (22–50 MPa)
Predict specific capacitance (F/g) of okra-derived activated carbons. Current density and configuration are top SHAP features.
Target: Specific_Capacitance_F_g (121–318 F/g)
Green synthesis of nanoparticles from okra. Model R² is negative, indicating high variability and insufficient data for reliable prediction.
Target: Inhibition_Zone_Diameter_mm — Predictor excluded
Missing experimental values were handled using Multivariate Imputation by Chained Equations (MICE), preserving inter-feature correlations across all five domains. Imputation improved dataset completeness without introducing uniform bias.
Three tree-based algorithms — Random Forest, Gradient Boosting, and Extra Trees — were trained independently per domain and combined via soft voting. This ensemble consistently outperformed individual models, particularly for adsorption (R²=0.8987) and electrochemistry (R²=0.8951).
SHapley Additive exPlanations (SHAP) decomposed each prediction into feature contributions. Key findings: corrosion current density (Icorr) is the dominant corrosion predictor; thermodynamic ΔH drives adsorption capacity; current density governs supercapacitor performance.
All R² and error metrics reported are from 5-fold stratified cross-validation, ensuring unbiased performance estimates for small experimental datasets (n = 6–67 per domain after imputation).
5-fold CV metrics. Bold ensemble rows represent the deployed predictor. R² values represent explained variance; MAE and RMSE are in target units (IE%, log(mg/g+1), MPa, F/g).
| Domain | Model | CV R² | CV MAE | CV RMSE |
|---|---|---|---|---|
| Corrosion Inhibition | Random Forest | 0.8424 | 5.364 | 10.859 |
| Gradient Boosting | 0.8226 | 5.309 | 11.521 | |
| Extra Trees | 0.8408 | 4.759 | 10.912 | |
| 🏆 Heterogeneous Voting | 0.8422 | 4.911 | 10.865 | |
| Adsorption / Pollutant Removal | Random Forest | 0.8774 | 0.350 | 0.468 |
| Gradient Boosting | 0.8841 | 0.301 | 0.455 | |
| Extra Trees | 0.8691 | 0.321 | 0.484 | |
| 🏆 Heterogeneous Voting | 0.8987 | 0.283 | 0.426 | |
| Biopolymers & Composites | Random Forest | 0.5591 | 2.880 | 4.113 |
| Gradient Boosting | 0.5735 | 2.906 | 4.046 | |
| Extra Trees | 0.5360 | 2.995 | 4.220 | |
| 🏆 Heterogeneous Voting | 0.5744 | 2.915 | 4.041 | |
| Electrochemistry / Carbon | Random Forest | 0.8146 | 12.231 | 17.598 |
| Gradient Boosting | 0.8693 | 10.123 | 14.772 | |
| Extra Trees | 0.8290 | 10.754 | 16.897 | |
| 🏆 Heterogeneous Voting | 0.8951 | 8.974 | 13.234 | |
| Nanoparticle Green Synthesis | Random Forest | −0.1999 | 1.178 | 2.017 |
| Gradient Boosting | −0.8028 | 1.430 | 2.472 | |
| Extra Trees | −0.7525 | 1.412 | 2.438 | |
| Heterogeneous Voting | −0.5268 | 1.292 | 2.275 |
Enter your experimental conditions across corrosion, adsorption, composites, or electrochemistry and receive an instant ML-based estimate with SHAP-guided insights.