Accurate　prediction　of　the　number　of　traffic　crashes　can　prejudge　road　traffic　safety　risks　and　fully　ensure　the　efficient　operation　of　the　transportation　system.　However,　existing　traffic　crash　prediction　is　based　on　data　from　cities　outside　China.　The　data　of　25,628　traffic　crashes　in　a　Chinese　megacity　from　2012　to　2013　were　collected,　and　a　traffic　crash　prediction　model　based　on　a　convolutional　neural　network　(CNN)　and　long　short　memory　network　(LSTM)　was　proposed.　Firstly,　the　feature　extraction　and　superposition　of　crash　data　were　carried　out　based　on　the　CNN　model.　Secondly,　the　LSTM　model　extracted　time　series　features　to　predict　traffic　crashes.　Finally,　the　CNN-LSTM　composite　model,　GM　(1,1),　ARIMA　model,　and　BP　neural　network　model　were　comprehensively　compared.　The　CNN-LSTM　combined　model　had　a　higher　prediction　accuracy　than　the　other　three.　The　research　conclusion　helps　better　determine　the　characteristics　and　laws　of　traffic　crashes　in　megacities　to　provide　a　reliable　decision-making　reference　for　traffic　management.