Abstract: This paper addresses the issue that fault currents in DC solid-state circuit breakers do not naturally pass through zero; if a fault is not detected in a timely manner, the fault current can rise rapidly and threaten equipment safety. To address this, an improved machine learning-based predictive method is proposed. The study uses simulation to generate data for various operating conditions, including normal operation, inrush current, sudden load increase, sustained overload, and short circuits, and combines information on current changes, fluctuations, and thermal accumulation to identify risks. Simulation results demonstrate that this method achieves a risk identification recall rate of 0.9834 and reduces the false negative rate to 0.0166, outperforming the comparison method. This indicates that the method can provide early warning and decision support for solid-state circuit breakers without compromising the rapid protection capabilities of the hardware.