The conventional treatment of heavy metal wastewater via chemical precipitation faces significant challenges, including the high cost of chemical agents and inconsistent treatment performance that risks regulatory non-compliance due to manual, sub-optimal dosing strategies. To address these limitations, this study proposes a novel, four-stage optimization framework that integrates Machine Learning (ML), Monte Carlo Simulation (MCS), and Particle Swarm Optimization (PSO) to determine the cost-optimized chemical dosages. Stage 1 utilizes a trained ML model to serve as the non-linear predictive constraint, accurately forecasting effluent quality. Stage 2 employs an uncertainty quantification module, comparing advanced sampling techniques like Latin Hypercube Sampling (LHS) to generate thousands of realistic operating scenarios, ensuring the solution is robust against real-world input variability. Stage 3 implements a penalized PSO algorithm to minimize the total chemical cost under the strict regulatory compliance constraint predicted by the ML model. Finally, Stage 4 develops a fast and efficient Surrogate Model trained on the optimal PSO results, enabling plant operators to obtain instantaneous, cost-optimized dosage recommendations for superior operational efficiency in a real-time environment. This framework successfully transforms deterministic dosing into a robust, cost-aware, and data-driven process, delivering significant economic savings while maintaining guaranteed treatment effectiveness.