Surface-normal electro-absorptive modulators based on III-V quantum well superlattices are of interest for a large number of applications, including on-chip and free-space photonic links, detection-and-ranging and optical tagging. In recent years, novel designs of the quantum well layers stack, such as asymmetric stepped quantum well, have allowed to reach energy efficient, wide spectral bandwidth modulation performance. Nevertheless, the design of these structures is still based on intuition rather than on a quantitative assessment of the device and system performance. Moreover, the increasing number of applications has highlighted the need for a comprehensive design approach, that incorporates the performance metrics of the specific system into the design considerations. We present a novel approach for the systematic optimization of the design of electro-absorptive modulators, based on a combination of analytical modeling and supervised machine learning. Fully-validated analytical modeling of the electronic transitions and optical propagation in the semiconductor compound is used for the training of an evolutionary algorithm, which drives the global search for optimal design. The approach was tested for the optimization of the superlattice design of the electro-absorptive modulator for two different applications: time-of-flight 3D ranging camera, and remote sensing of electro-chemical signal via optical tagging. In both cases, a system-specific figure-of-merit is proposed and employed for the evaluation and optimization of the performance, yielding two novel optimized designs which allow for considerable performance improvement of the respective systems.