Microwave-assisted pyrolysis (MAP) represents a promising sustainable energy technology for converting plastic waste into value-added fuels and chemicals. However, accurate prediction of product distribution remains challenging due to complex nonlinear interactions among feedstock composition, operating parameters, and absorber characteristics. This study uniquely integrates absorber (susceptor) type with feedstock composition and operating parameters within a multi-stage, explainable machine learning (ML) framework for MAP of waste plastics. A digital intelligence approach based on ensemble tree-based regression and explainable artificial intelligence is developed to enable multi-target prediction of gas, liquid, and wax yields, along with oil higher heating value (HHV). A three-stage modeling strategy was constructed using a 26-feature dataset incorporating elemental composition, process conditions, and absorber type. Among the evaluated models, XGBoost achieved the highest predictive performance, with R2 values of 0.83 for gas yield and 0.91 for oil HHV, along with notable reductions in MAE and RMSE upon inclusion of absorber features. SHapley Additive exPlanations (SHAP) and partial dependence analyses provided mechanistic interpretability, identifying temperature, absorber type, hydrogen content, and volatile matter as dominant factors governing product selectivity. Absorbers such as SiC, activated carbon, Fe-based materials, and graphite were shown to enhance secondary cracking and vapor upgrading. The proposed absorber-integrated, explainable learning framework demonstrates the potential of digital intelligence for optimizing waste-to-energy systems, improving process reliability, and supporting sustainable fuel production within circular economy pathways.