Historical　tourism　volume,　search　engine　data,　and　weather　calendar　data　have　close　causal　relationship　with　daily　tourism　volume.　However,　when　used　in　the　prediction　of　daily　tourism　volume,　the　feature　variables　of　the　huge　and　complex　search　engine　data　do　not　have　strong　independence.　These　repetitive　and　highly　relevant　data　must　be　analyzed　and　selected;　otherwise,　they　will　increase　the　training　burden　of　neural　network　and　reduce　the　prediction　effect.　This　study　proposes　a　daily　tourism　volume　prediction　model,　maximum　correlation　minimum　redundancy　feature　selection　and　long　short-term　memory,　on　the　basis　of　feature　selection　and　deep　learning.　Firstly,　the　multivariate　high-dimensional　features,　including　search　engine　data　and　weather　factors,　are　selected　to　identify　the　key　influencing　factors.　Secondly,　the　deep　neural　network　is　used　to　make　a　multistep　forward　rolling　prediction　of　daily　tourism　volume.　Results　show　that　keywords　of　famous　scenic　spots,　weather,　historical　tourism　volume,　and　tourism　strategies　in　the　search　engine　data　significantly　improve　the　prediction　accuracy　of　daily　tourism　volume.　The　proposed　maximum　correlation　minimum　redundancy　feature　selection　and　long　short-term　memory　model　performs　better　than　other　models,　such　as　autoregressive　integrated　moving　average,　multiple　regression,　support　vector　machine,　and　long　short-term　memory.