Scientific　investment　forecasting　can　effectively　avoid　the　blind　investments　of　environmental　management.　Among　existing　studies　in　developing　investment　forecasting　models,　the　extended　belief　rule-based　system　(EBRBS)　showed　its　potential　to　accurately　predict　environment　investments　but　also　exposed　two　challenges　to　be　further　addressed:　(1)　how　to　select　antecedent　attributes　from　various　environmental　indicators　for　the　EBRBS;　(2)　how　to　optimize　basic　parameters　of　the　EBRBS　based　on　the　selected　antecedent　attributes.　Since　these　two　challenges　are　connected,　a　bi-level　joint　optimization　model　is　proposed　to　improve　the　EBRBS　for　better　environmental　investment　forecasting,　in　which　the　selection　of　antecedent　attributes　is　described　as　an　upper-level　optimization　model　using　Akaike　information　criterion　(AIC)　and　the　optimization　of　basic　parameters　is　as　a　lower-level　optimization　model　using　mean　absolute　error　(MAE).　Moreover,　a　corresponding　bilevel　joint　optimization　algorithm　is　proposed　to　solve　the　bi-level　joint　optimization　model,　where　ensemble　feature　selection　and　swarm　intelligence　optimization　are　regarded　as　the　engine　of　upperlevel　and　lower-level　optimizations,　respectively.　The　real　environmental　data　collected　from　2005　to　2020　of　30　Chinese　provinces　are　studied　to　verify　the　effectiveness　of　the　proposed　approach.　Experimental　results　show　that　the　EBRBS　with　bi-level　joint　optimization　not　only　can　effectively　predict　environmental　investments,　but　also　is　able　to　have　desired　accuracy　better　than　previous　investment　forecasting　models.