Unpredictable　stock　market　factors　make　it　difficult　to　predict　stock　index　futures.　Although　efforts　to　develop　an　effective　prediction　method　have　a　long　history,　recent　developments　in　artificial　intelligence　and　the　use　of　artificial　neural　networks　have　increased　our　success　in　nonlinear　approximation.　When　we　study　financial　markets,　we　can　now　extract　features　from　a　big　data　environment　without　prior　predictive　information.　We　here　propose　to　further　improve　this　predictive　performance　using　a　combination　of　a　deep-learning-based　stock　index　futures　prediction　model,　an　autoencoder,　and　a　restricted　Boltzmann　machine.　We　use　high-frequency　data　to　examine　the　predictive　performance　of　deep　learning,　and　we　compare　three　traditional　artificial　neural　networks:　1)　the　back　propagation　neural　network;　2)　the　extreme　learning　machine;　and　3)　the　radial　basis　function　neural　network.　We　use　all　of　the　1-min　high-frequency　transaction　data　of　the　CSI　300　futures　contract　(IF1704)　in　our　empirical　analysis,　and　we　test　three　groups　of　different　volume　samples　to　validate　our　observations.　We　find　that　the　deep　learning　method　of　predicting　stock　index　futures　outperforms　the　back　propagation,　the　extreme　learning　machine,　and　the　radial　basis　function　neural　network　in　its　fitting　degree　and　directional　predictive　accuracy.　We　also　find　that　increasing　the　amount　of　data　increases　predictive　performance.　This　indicates　that　deep　learning　captures　the　nonlinear　features　of　transaction　data　and　can　serve　as　a　powerful　stock　index　futures　prediction　tool　for　financial　market　investors.