Contest　swings　are　influenced　by　multiple　and　complex　factors,　which　makes　accurate　prediction　a　challenging　task.　This　study　aims　to　utilize　the　advanced　LSTM　recurrent　neural　network　with　the　random　forest　algorithm　to　make　predictions　of　race　swings　and　analyze　the　related　important　factors　in　depth.　The　experimental　results　show　that　the　LSTM　recurrent　neural　network　can　effectively　capture　the　swings　occurring　in　the　contest　and　the　high　accuracy　of　the　model　is　confirmed　through　relevant　metrics.　In　addition,　the　analytical　results　of　the　random　forest　model　not　only　determine　the　importance　of　each　factor　to　the　fluctuation　of　the　competition　but　also　propose　corresponding　coping　strategies,　which　provide　an　important　reference　basis　for　further　coping　with　the　fluctuation.　These　findings　provide　important　empirical　support　for　a　deeper　understanding　of　the　complexity　of　race　swings　and　the　development　of　effects.　　©　2024　IEEE.