In　recent　years,　the　importance　of　user　information　has　increased　rapidly　for　context-aware　applications.　This　paper　proposes　a　deep　learning　mechanism　to　identify　the　transportation　modes　of　smartphone　users.　The　proposed　mechanism　is　evaluated　on　a　database　that　contains　more　than　1000　h　of　accelerometer,　magnetometer,　and　gyroscope　measurements　from　five　transportation　modes,　including　still,　walk,　run,　bike,　and　vehicle.　Experimental　results　confirm　the　effectiveness　of　the　proposed　mechanism,　which　achieves　approximately　95%　classification　accuracy　and　outperforms　four　well-known　machine　learning　methods.　Meanwhile,　we　investigated　the　model　size　and　execution　time　of　different　algorithms　to　address　practical　issues.　©　2001-2012　IEEE.