Proactive decision-making is crucial for the safety and comfort of intelligent vehicles in complex, human-dominated traffic. However, traditional reactive controllers and decoupled prediction-decision pipelines often fail to mitigate sudden risks like dangerous cut-ins. To address this, we propose a Predictive Decision and Control (PDC) framework that tightly couples long-horizon behavior prediction with a risk-aware control module. The framework features a heterogeneous prediction module that integrates a Long Short-Term Memory (LSTM) network for multi-step trajectory prediction and a Least-Squares Gradient Boosting (LSBoost) model for lane-change intent recognition. This predictive information is then translated into a proactive safety reference speed by a risk-aware algorithm that explicitly evaluates potential cut-in threats. Through co-simulations in hazardous scenarios, our PDC framework demonstrated significant improvements over a standard Adaptive Cruise Control (ACC) baseline. Notably, it improved the minimum Time-to-Collision (TTC) by 191%, while also reducing mean absolute acceleration by 18.3% and energy consumption by 18.7%. The framework effectively translates long-horizon predictions into proactive control actions, enhancing active safety, ride comfort, and energy efficiency.