Abstract: With the continuous development of industries such as integrated circuits, wind power, and nuclear energy, the accumulation of spent thermosetting resin-based composites has emerged as an increasingly pressing environmental issue. Pyrolysis represents a promising technology for the resource recovery and value-added utilization of these wastes. To elucidate the pyrolysis characteristics of such wastes, this study systematically investigated the thermal decomposition behavior of spent ion-exchange resins based on a styrene-divinylbenzene backbone functionalized with sodium sulfonate groups. In addition, artificial intelligence models were developed to predict key pyrolysis parameters across different types of thermosetting resin-based composite wastes. The mass-loss behavior and heat flow evolution during pyrolysis were analyzed using thermogravimetry-differential scanning calorimetry (TG-DSC). The composition and distribution of gaseous and liquid products were further characterized by thermogravimetry-mass spectrometry (TG-MS) and pyrolysis-gas chromatography/mass spectrometry (Py-GC/MS). The results indicate that the cleavage of the styrene-divinylbenzene crosslinked backbone occurred predominantly between 415 and 505 °C. During this stage, the major pyrolysis products were styrene, ethylbenzene, and toluene—high-value chemicals that accounted for approximately 77% of the detected products at 455 °C. At temperatures above 581 °C, CH₄, H₂, and CO₂ became the dominant gaseous products, forming combustible gases with potential for energy recovery, while a char yield of approximately 45% was observed. An increase in heating rate led to a higher temperature corresponding to the maximum mass-loss rate, a broader temperature range for backbone cleavage, and a higher overall mass-loss rate. These changes collectively influenced the temperature window and yield of volatile products as well as the amount of residual char. Therefore, the heating rate is a key process parameter for the efficient recovery of gas, liquid, and char products from spent ion-exchange resins. Furthermore, regression models and artificial neural network (ANN) models were developed by integrating experimental results from this study with literature data on various thermosetting resin-based wastes. Based on feature importance analysis using the F-test, these models were trained using the proximate and ultimate analyses of spent resins to predict their pyrolysis parameters, including onset temperature, temperature of maximum mass-loss rate, termination temperature, and overall weight loss. Among all modeling approaches, the ANN trained using the Levenberg-Marquardt algorithm exhibited the best predictive performance, achieving a coefficient of determination (R²) of 0.99 and a mean squared error of 0.0007. In future research, emphasis should be placed on improving the purity of gaseous and liquid products, enhancing the performance of char materials, expanding the experimental database, and employing more advanced machine learning techniques. These efforts will further improve the generalization and predictive accuracy of models, thereby providing more reliable guidance for optimizing pyrolysis processes toward the efficient synergistic recovery of gas, liquid, and char products from thermosetting resin-based composite wastes.