Optimizing machining parameters for Ti-6Al-4V turning remains challenging due to conflicting objectives between surface quality and productivity. This study presents an integrated framework combining deep neural network regression (DNNR) with multi-objective evolutionary optimization to simultaneously predict and optimize surface roughness (Ra) and material removal rate (MRR). A Box-Behnken design with 15 experimental runs was conducted using three cutting parameters: cutting speed (40–60 m/min), feed rate (0.05–0.2 mm/rev), and depth of cut (0.2–1.2 mm). The developed DNNR model demonstrated exceptional predictive accuracy with R2 values of 0.934 for Ra and 0.998 for MRR, validating its effectiveness as a surrogate model. Subsequently, the Non-dominated Sorting Genetic Algorithm II (NSGA-II) was employed to generate a Pareto front of 33 optimal solutions, revealing critical trade-offs between surface finish and productivity. Sensitivity analysis identified cutting speed as the most influential parameter, followed by feed rate with pronounced nonlinear effects. To facilitate decision-making, entropy-based TOPSIS ranking was applied to the Pareto solutions, yielding optimal cutting parameters of Vc = 40.092 m/min, fz = 0.097 mm/rev, and ap = 0.210 mm with predicted Ra = 0.021 μm and MRR = 0.017 cm3/min. This top-ranked solution achieved a TOPSIS score of 1.000, representing the ideal balance between minimal surface roughness and acceptable productivity. The proposed DNNR-NSGA-II-TOPSIS framework significantly reduces experimental requirements while providing reliable optimization for Ti-6Al-4V turning operations, making it suitable for aerospace and biomedical manufacturing applications.