Dioxin　(DXN)　is　a　kind　of　persistent　organic　pollutant　with　a　cumulative　effect.　It　is　also　one　of　the　main　reasons　for　＇not　in　my　back　yard＇　effect　in　Municipal　solid　waste　incineration　(MSWI)　plants.　Real-time　detection　of　DXN　is　helpful　to　realize　emission　reduction,　optimize　control,　and　eliminate　oppose　effect　in　MSWI　process.　However,　there　are　very　tiny　label　process　data　that　can　be　used　to　construct　data-driven　prediction　models　due　to　the　time　and　economic　cost.　In　order　to　utilize　the　process　data,　this　article　presents　a　collaborative　training　decision　trees　(CTDTs)　method　for　dioxin　emission　concentration　prediction.　First,　the　raw　label　process　data　is　used　to　train　the　decision　tree　model,　after　that　the　process　data　is　labeled.　Second,　the　root　mean　square　error　of　the　labeled　sample　is　calculated　to　select　the　optimal　labeled　and　process　data.　Third,　the　DXN　emission　prediction　model　is　constructed　by　cross-combination　of　the　raw　labels　and　labeled　process　data.　Simulation　results　of　the　benchmark　dataset　and　practical　DXN　data　verify　the　effectiveness　of　the　proposed　method.　©　2021　IEEE.