For　mobile-edge　computing　(MEC),　a　machine　learning-based　stochastic　task　offloading　algorithm　was　proposed.　By　dividing　the　task　into　offloadable　components　and　unoffloadable　components,　the　improved　Q　learning　and　deep　learning　algorithm　were　used　to　generate　the　optimal　offloading　strategy　of　stochastic　task,　which　minimized　the　weighted　sum　of　energy　consumption　and　time　delay　of　the　mobile　devices.　The　simulation　results　show　that　the　proposed　algorithm　saves　the　weighted　sum　of　energy　consumption　and　time　delay　by　38.1%,　compared　to　the　local　execution　algorithm.　©　2019,　Editorial　Department　of　Journal　of　Beijing　University　of　Posts　and　Telecommunications.　All　right　reserved.