Proximity-based interference is a growing threat to satellite communications, driven by dense multi-orbit constellations and increasingly agile adversarial maneuvers. We propose a hybrid simulation framework that integrates orbital maneuver modeling with RF signal degradation analysis to detect and classify suspicious proximity operations. Using the open-source Maneuver Detection Data Generation (MaDDG) library from MIT Lincoln Laboratory, we generate labeled datasets combining impulsive maneuver profiles with radio-frequency (RF) impacts across a range of behavioral intents: routine station-keeping, covert shadowing, and overt jamming. Our approach fuses kinematic features such as range, velocity, acceleration, and Time of Closest Approach (TCA), with RF metrics including Received Signal Strength Indicator (RSSI), throughput, and Jammer-to-Signal Ratio (JSR). These features are further enhanced with temporal derivatives and rolling-window statistics to capture subtle or transient interference patterns. A Random Forest classifier trained on this fused feature set achieves 94.67% accuracy and a macro F1 score of 0.9471, outperforming models using only kinematic or RF inputs. The system is particularly effective in detecting covert threats, such as surveillance or intermittent jamming, that evade RF-only methods.