In this work, a heterostructure photodetector that uses few-layer graphene (FLG) is proposed for ultra-fast near-infrared (NIR) optoelectronic applications. The device has a vertically stacked FLG/InSb/AlSb structure, built on an n-type GaAs substrate. In this configuration, the FLG acts as a high-mobility conductive contact, thereby facilitating efficient carrier transport, while the narrow bandgap InSb layer serves as the primary light-absorbing region for NIR radiation. Furthermore, the AlSb layer acts as a barrier that improves carrier confinement and promotes charge separation at the heterojunction interface. When exposed to NIR illumination, the electron-hole pairs are mainly produced in the InSb absorption layer, which are then separated by the built-in electric field, resulting in effective photocurrent generation. The high carrier mobility and ballistic transport properties of FLG contribute to rapid carrier collection and improved photoresponse. The device’s performance is evaluated by employing Silvaco ATLAS TCAD simulation software. The proposed graphene-based heterostructure photodetector exhibits stable current-voltage characteristics under illumination, improved photocurrent density with a photocurrent-to-dark current density ratio of 1010, responsivity of 0.53 A/W, detectivity of 2.4 × 1016 cmHz1/2/W, and ultrafast response characteristics with a rise (fall) time of 0.51 (0.54) ns. Moreover, machine learning regression models, such as Random Forest, Extra Trees, XGBoost, and CatBoost, are used on the simulated dataset to learn how device parameters affect key performance metrics. This makes it possible to accurately predict and efficiently optimize the characteristics of photodetectors at a lower computational cost. These results demonstrate the potential of the proposed heterostructure for high-speed and high-sensitivity NIR applications, such as optical communication, environmental monitoring, and infrared imaging systems.