Machine Learning and Hybrid Fuzzy Logic for Prediction and Optimization of Flexible Supercapacitor Performance for Smart Textiles

In this study, a hybrid decision support system combining Fuzzy AHP, Type-2 Fuzzy TOPSIS, and XGBoost+SHAP methods has been developed for performance optimization of smart textile-based flexible supercapacitor electrodes. Experimental data were collected under over 250 different synthesis conditions to investigate the effects of synthesis parameters; surface area, pore size, conductivity, film thickness, synthesis temperature, flexibility on performance outputs; specific capacitance, energy density, cyclic stability. The highest value for specific capacitance (912 F/g) was obtained at surface area 2450 m²/g, pore size 22 nm, conductivity 8500 S/cm and synthesis temperature 1050°C. According to Fuzzy AHP, the weight of surface area was 0.68, while XGBoost+SHAP showed that this parameter provides an average positive contribution of +47 F/g. The maximum energy density (72.4 Wh/kg) was measured at synthesis temperature 1200°C, conductivity 9800 S/cm, surface area 2300 m²/g. In Type-2 TOPSIS, this point was classified as "excellent" (closeness coefficient 0.94). The highest cyclic stability (92.3%) was observed at pore size 35 nm, film thickness 125 µm, synthesis temperature 700°C. SHAP analysis showed that stability decreases by 12% when pore size falls below 25 nm. High agreement was found between the prediction accuracy of the hybrid model (R² = 0.94) and experimental results.