This work proposes a machine learning (ML) assisted multi-objective optimization (MOO) approach for reverse engineering of vinyl acetate polymerization processes. The method leverages ML models trained on data from kinetic Monte Carlo (kMC) simulations to minimize number of expensive laboratory experiments. A genetic algorithm (GA) as the MOO optimizer is employed, considering reaction time, monomer conversion, and molar mass distribution (MMD) similarity as objectives. The trained ML models assist the optimization process and predict key polymer properties for candidate recipes generated by the GA, enabling rapid fitness function evaluation and allow for producing novel previously not simulated polymerization recipes. The proposed framework involves: (1) training ML models for monomer concentration and MMD prediction using kMC simulation data; (2) performing GA-based MOO to identify optimal recipes (Pareto front) for a target MMD (3) selecting the most suitable recipe based on user priorities from the resulting Pareto front, considering user-defined weights for each objective (time, conversion, MMD). The experiments demonstrate that the GA, coupled with simulation-supported ML, efficiently identifies optimal recipes with high accuracy. Notably, the ML models achieve good performance even with limited training data. This approach offers a rapid and cost-effective solution for reverse engineering of vinyl acetate polymerization processes.