Numerous　investigations　have　shown　that　the　municipal　solid　waste　incineration　(MSWI)　has　become　one　of　the　major　sources　of　dioxin　(DXN)　emissions.　Currently,　the　primary　issue　that　needs　to　be　addressed　for　DXN　emission　reduction　control　is　the　online　measurement　of　DXN.　Data-driven　AI　algorithms　enable　real-time　DXN　concentration　measurement,　facilitating　its　control.　However,　researchers　mainly　focus　on　building　models　for　DXN　emissions　at　the　stack.　This　approach　does　not　allow　for　the　construction　of　models　that　online　measurement　of　DXN　generation　and　absorption　throughout　the　whole　process.　To　achieve　optimal　pollution　control,　models　that　encompass　the　whole　process　are　necessary,　not　just　models　focused　on　the　stack.　Therefore,　this　article　focuses　on　modeling　the　whole　process　of　DXN　concentrations,　including　generation,　adsorption,　and　emission.　It　uses　machine　learning　techniques　based　on　advanced　tree-based　data-driven　deep　and　broad　learning　algorithms.　The　determination　of　data　characteristics　at　different　phases　is　grounded　in　the　understanding　of　the　DXN　mechanism,　offering　a　novel　framework　for　DXN　modeling.　State-of-the-art　tree-based　models,　including　adaptive　deep　forest　regression　algorithm　based　on　cross　layer　full　connection,　tree　broad　learning　system,　fuzzy　forest　regression,　and　aid　modeling　technologies,　are　applied　to　handle　diverse　data　characteristics.　These　characteristics　encompass　high-dimensional　small　samples,　low-dimensional　ultra-small　size　samples,　and　medium-　dimensional　small　samples　across　different　phases　related　to　DXN.　The　most　interesting　is　the　robust　validation　where　the　proposed　a　whole　process　tree-based　model　for　DXN　is　validated　using　nearly　one　year　of　authentic　data　on　DXN　generation,　adsorption,　and　emission　phases　in　an　MSWI　plant　of　Beijing.　The　proposed　modeling　framework　can　be　used　to　explore　the　mechanism　characterization　and　support　the　pollution　reduction　optimal　control.