Due to inevitable external mechanical abusive loadings, lithium-ion batteries (LIB) will suffer damages or defects. If the safety risk level of the cell is unknown, an early decision cannot be made. In this work, we develop a Random Forest (RF) based classification model to implement online safety risk level classification with low time cost and high accuracy. Four levels of battery cell safety risk are defined: a). Normal; b), Latent risk (defective cells, short circuit but normal operation); c). Low risk (short circuit without possible TR triggering); d). High risk (short circuit with possible TR triggering). The training dataset combines experimental data and simulation data from the multi-physical model. The training samples consist of voltage, current, and temperature signals under different operating conditions and different risk level scenarios. The prediction results show that the classifiers have a good performance and robustness. This approach will provide insight into the identification, monitoring, and early warning of battery safety issues.