The optimization of rotor profiles in screw compressors plays a crucial role in enhancing overall efficiency and performance under different operating conditions. Conventional rack generation methods serve as foundational tools for manipulating rotor profiles and ensuring the correct kinematics. However, to further explore the design space of screw profiles, machine learning (ML) can be leveraged and integrated to couple geometry manipulation and compressor performance estimation. To this end, the rack profile has been parametrized into segments based on normal and position vectors to allow the systematic generation of diverse profile geometries based on imposed constraints and/or objective functions. To conduct a global search, a genetic algorithm (GA) has been modified to include appropriate fitness criteria to automate the evolutionary profile generation process. Candidate profiles are evaluated using an ML model trained with a 1D compressor mechanistic model to predict performance metrics and both flow and mechanical losses. This integrated approach significantly reduces the computational time for completing a full GA optimization. The algorithm was trained and executed entirely on CPU due to the logic-intensive nature of geometry evaluation and performance prediction, which favor structured decision-making over the highly parallel workloads typically suited for GPU acceleration. An oil-injected screw compressor for industrial applications has been used as the case study. The results from the optimization runs are discussed and compared with higher-fidelity models such as CFD and a fully mechanistic compressor model.