Data-driven valorization of mixed olefins byproducts via self-stabilized precipitation polymerization

The transition of fossil fuels from energy sources to chemical feedstocks has generated huge amount of mixed olefins byproducts, whose efficient valorization remains an important challenge in chemical engineering. Self-stabilized precipitation (2SP) polymerization offers a promising route to directly convert these under-utilized olefins into polymer products without costly separation or other pre-treatments. However, the multi-factor nature of 2SP leaves its mechanism largely speculative, and the reaction feasibility currently relies on time-consuming trial-and-error experimentation. Here, we developed a machine learning predictor, 2SP-MLP, trained on 844 experimental data points to enable rapid and accurate prediction of 2SP outcomes. Following Bayesian optimization with 5-fold cross-validation, the XGBoost classifier achieved the best performance (ACC = 0.90, AUC = 0.96). Interpretability analysis identified the monomer-solvent solubility parameter difference (exceeding 3.6 MPa 1/2 ) as the key factor governing precipitation feasibility. Beyond solubility parameter, mechanistic analyses suggested that molecular shape, flexibility and charge distribution influenced polymer precipitation and microparticle formation (for polymer precipitation, ACC = 0.94 and AUC = 0.97; for microparticle formation, ACC = 0.67 and AUC = 0.71). Regression models for particle yield and olefin conversion both performed best with XGBoost (R 2 = 0.73 and 0.89, respectively). Experimental validation on 10 untested formulations with different monomer combinations, feed ratios, and solvent types achieved 70% prediction accuracy, especially on FT mixed olefins-related systems. Overall, this work provides a data-driven tool that accelerates process design and reduces experimental cost for 2SP-based valorization of mixed olefins, demonstrating the potential of ML to advance sustainable chemical manufacturing.