Opportunistic　communications　as　an　efficient　traffic　offloading　method　can　be　used　to　offload　uplink　traffic　of　cellular　networks　to　Wi-Fi　networks.　However,　because　of　its　contact　pattern　(contact　frequency　and　contact　duration)　the　offloading　method　could　not　ensure　the　data　to　be　successfully　offloaded　to　Wi-Fi　Access　Points　(APs)　within　a　time　constraint.　In　this　paper,　we　focus　on　maximizing　the　probability　of　offloading　data　to　Wi-Fi　APs　by　fragmenting　the　data　and　assigning　the　fragments　to　different　direct　or　indirect　paths　generated　by　opportunistic　contacts.　Firstly,　we　propose　two　methods　based　on　mobility　prediction,　which　is　realized　by　machine　learning,　to　separately　calculate　the　probability　of　offloading　data　to　Wi-Fi　APs　by　the　direct　offloading　path　considering　multiple　opportunistic　contacts　and　contact　duration,　and　the　probability　of　indirectly　offloading　data　to　Wi-Fi　APs　by　the　indirect　offloading　path.　Then,　based　on　the　probability　calculation　methods　the　offloading　probability　maximization　is　formulated　as　a　non-linear　integer　programming　problem,　and　we　propose　a　distributed　heuristic　algorithm　to　solve　it　considering　complexity　of　the　probability　calculation　and　limited　computation　capacities　of　devices.　Simulation　results　prove　the　data　offloading　probability　of　our　proposed　algorithm　outperforms　other　algorithms　under　different　simulation　environment.