Sustainable valorization of mixed Saudi coffee waste and polystyrene via co-pyrolysis: Thermokinetic analysis and machine learning-based modelling using DNN and kNN

In accordance with Saudi Arabia’s Vision 2030 and global initiatives to address plastic pollution through circular economy strategies, this study explores the thermal co-pyrolysis of recycled Saudi coffee (RSC) biomass and waste polystyrene (PS) plastic blends. Employing thermogravimetric analysis (TGA) at heating rates of 5, 10, and 20 °C·min − 1, the investigation analyses the synergistic effects, kinetics, and thermodynamics of RSC/PS blends. The findings demonstrate that while pure RSC degrades within the temperature range of 226–386 °C and pure PS decomposes between 290 and 484 °C, the blending of these materials induces a positive synergy, resulting in a reduction of PS bond energy and the emergence of distinct two-stage DTG peaks. Kinetic analysis (10–90% conversion) shows that increasing RSC content from 25% to 75% shifts the pyrolysis mechanism from diffusion-controlled (D6 → D3) to surface-reaction controlled (F3). This shift stems from increased biochar formation, which reduces the molten polymer phase. According to five isoconversional models—FWO-St, FWO-It, KAS, Vyazovkin, and Šesták–Berggren – the 50%RSC blend is the most reactive, boasting the lowest activation energy (87–92kJ·mol −1 ) and highest efficiency. Conversely, the 75%RSC blend follows a more stable surface-reaction process with a higher activation energy range of 178–187 kJ·mol −1 . Thermodynamic analysis reveals that the formation of the activated complex is endothermic, non-spontaneous, and stable. Machine learning approaches were employed to model the process, with a Deep Neural Network (DNN, 30–30-40, R 2 ≈ 0.990 ) and a k-Nearest Neighbors (kNN, k = 100, Chebyshev distance) model demonstrating high predictive accuracy. kNN provided superior fitting ( R 2 ≈ 1.000 ), with an RMSE of 0.530 and R 2 ≈ 1.000 . Pearson correlation identified temperature ( r = - 0.78 ) as the primary driver of mass loss. This study demonstrates that co-pyrolysis of RSC/PS is a feasible, energy-efficient route for waste valorization, with 50/50 blending recommended for industrial reactors governed by machine learning techniques.