Rubber vibration isolators are widely used in various power systems and mechanical equipment of ships. However, conventional rubber vibration isolator performance prediction models and design methods exhibit high computational costs in structural optimization and design cycles. To address these issues, this study proposes a data-driven approach for rubber vibration isolator performance prediction and structural optimization design. A machine learning model, artificial neural network (ANN), is proposed to replace the FEM model to predict the static behavior of the rubber vibration isolator. Combined with the gradient descent optimization algorithms, a reverse design approach is proposed to realize the rubber vibration isolator design. The reverse design process based on this framework shortens the optimization design cycle of a single case by 90% compared with the optimization process based on the FEM method. This research provides an innovative solution for the intelligent design of rubber vibration isolation devices, effectively bridging the gap between traditional empirical design and performance optimization requirements.