<p>This study develops a comprehensive and practical optimization workflow for trochoidal milling of AISI 4140 steel under dry conditions. The main objective is to simultaneously reduce surface roughness (<i>Ra</i>), tool wear (<i>VB</i>), and cutting forces (<i>Fc</i>), while increasing the material removal rate (<i>MRR</i>). A systematic series of experiments was conducted using a D-optimal design, with controlled variations of key parameters, including cutting speed (<i>Vc</i>), feed per tooth (<i>fz</i>), axial depth of cut (<i>ap</i>), and step-over (<i>St</i>). Advanced machine learning models, including Kolmogorov–Arnold Networks (KAN), CatBoost (CAT), Gradient Boosting Regressor (GBR), LightGBM (LGB), and XGBoost (XGB), were trained using Bayesian optimization to predict machining outcomes accurately. These models were then used as initial functions for the NSGA-III algorithm to generate a Pareto front of optimal multi-objective solutions. To select the best parameter set from the Pareto solutions, the study applied multi-criteria decision analysis (MCDA), combining ranking methods such as CRADIS, COPRAS, ARAS, and MARCOS, together with objective weighting techniques including CRITIC and GINI. Experimental validation showed that the prediction error for the entire workflow was below 8.5% for all objectives. This framework represents a truly data-driven approach to parameter optimization, combining prediction, optimization, and objective selection in a transparent and systematic manner. The results not only outperform traditional methods but also provide a practical decision-support tool that can be applied effectively in real milling processes for smart and sustainable manufacturing.</p>