Low-Heating-Rate Thermal Degradation of Date Seed Powder and HDPE Plastic: Machine Learning CDNN, MLRM, and Thermokinetic Analysis.

Finding reliable, sustainable, and economical methods for addressing the relentless increase in plastic production and the corresponding rise in plastic waste within terrestrial and marine environments has garnered significant attention from environmental organizations and policymakers worldwide. This study presents a comprehensive analysis of the low-heating-rate thermal degradation of high-density polyethylene (HDPE) plastic in conjunction with date seed powder (DSP), utilizing thermogravimetric analysis coupled with Fourier transform infrared spectroscopy (TGA/FTIR), machine learning convolutional deep neural networks (CDNNs), multiple linear regression model (MLRM) and thermokinetics. The TGA/FTIR experimental measurements indicated a synergistic interaction between the selected materials, facilitated by the presence of hemicellulose and cellulose in the DSP biomass. In contrast, the presence of lignin was found to hinder degradation at elevated temperatures. The application of machine learning CDNNs facilitated the formulation and training of learning algorithms, resulting in an optimized architectural composition comprising three hidden neurons and employing 27,456 epochs. This modeling approach generated predicted responses that are closely aligned with experimental results (R2 ~ 0.939) when comparing the responses from a formulated MLRM model (R2 ~ 0.818). The CDNN models were utilized to estimate interpolated thermograms, representing the limits of experimental variability and conditions, thereby highlighting temperature as the most sensitive parameter governing the degradation process. The Borchardt and Daniels (BD) model-fitting and Kissinger-Akahira-Sunose (KAS) model-free kinetic methods were employed to estimate the kinetic and thermodynamic parameters of the degradation process. This yielded activation energy estimates ranging from 40.419 to 91.010 kJ·mol⁻