To　address　the　existing　issue　of　electric　shock　incidents　that　cannot　be　accurately　identified　by　current　leakage　protection　devices,　this　paper　presents　a　novel　electric　shock　accident　recognition　method.　Firstly,　the　method　of　singular　spectrum　analysis　(SSA)　is　employed　to　extract　the　main　components　of　leakage　recording　data.　Subsequently,　20　temporal　domain　features　of　the　leakage　current　waveform　are　extracted.　Then,　an　ensemble　learning　model　based　on　extreme　gradient　boosting　(XGBoost),　categorical　boosting　(CatBoost)　and　random　forest　(RF),　is　established　to　select　optimal　features　that　best　represent　the　sample　characteristics　from　the　feature　set.　Finally,　support　vector　machine　(SVM)　is　used　to　classify　the　extracted　dataset.　Experimental　results　demonstrate　that　this　method　can　rapidly　differentiate　between　electric　shock　faults　and　common　leakage　faults,　achieving　an　accuracy　rate　as　high　as　99%,　indicating　its　feasibility.　©　2024　IEEE.