<p>Machine learning enhanced novel workflow is employed to determine the process parameters of fused filament fabrication (FFF) for optimizing the mechanical properties of Acrylonitrile butadiene styrene (ABS). This workflow integrates the Taguchi L-9 orthogonal array experimental design with machine learning methods such as symbolic regression (SR) and deep neural networks (DNN) to develop empirical relationships between identified process parameters (bed temperature, layer thickness, printing speed, and nozzle temperature) to the mechanical properties (tensile strength, elongation at break, toughness, fracture toughness, flexure modulus, and flexural strength). The experimentally generated data is used to effectively develop empirical principles encapsulated by terms in the SR-derived mathematical formulas. Later these formulas are feature-engineered using DNN before the targeted properties are optimized using non-dominated sorting genetic algorithm-II (NSGA-II). Based on SR approach, layer thickness appeared to have a significant impact on four of six mechanical properties of ABS, whereas bed temperature has the least impact on all mechanical properties. The DNN predictions further improved the SR results with enhanced R² values ranging from 0.96 to 0.997. Finally, multi-objective optimization pipeline of NSGA-II predictions of optimal layer thickness aligns with the lowest experimentally tested value (0.15&#xa0;mm), the optimal printing speed is close to the mid-range value (50&#xa0;mm/s), and the optimal bed and nozzle temperatures nearly match the experimental maxima of 105&#xa0;°C and 260&#xa0;°C. This hybrid machine learning workflow philosophy could be adapted to relate process parameters to mechanical properties not only for other 3D-printed materials but also for traditionally fabricated materials.</p> Graphical Abstract <p></p>