Diabetic　retinopathy　(DR)　is　one　of　the　complications　of　diabetes　mellitus,　which　is　an　important　manifestation　of　diabetic　microangiopathy　and　major　cause　of　vision　loss　in　middle-aged　and　elderly　people　worldwide.　Establishing　a　risk　prediction　model　for　diabetic　retinopathy　can　discover　high-risk　groups　and　early　warn　diabetic　retinopathy,　which　can　effectively　reduce　the　medical　cost　of　diabetes.　The　experimental　data　was　derived　from　the　electronic　medical　records　of　a　tertiary　hospital　of　Beijing　from　2013　to　2017,　including　29　inspection　indicators.　In　this　study,　we　compared　the　predictive　models　of　type　2　diabetes　mellitus　complicated　with　retinopathy,　and　finally　selected　the　random　forest　method　to　construct　the　risk　prediction　model.　The　weights　of　each　index　are　analyzed　by　linear　regression　algorithm,　the　combination　of　inspection　indicators　with　the　highest　accuracy　is　selected,　and　the　random　forest　model　is　optimized　to　improve　the　accuracy　of　the　classification　prediction　model,　accuracy　increased　by　3.7264%.　The　predictive　model　provides　a　basis　for　early　diagnosis　of　diabetic　retina　and　optimization　of　the　diagnostic　process.　©　2019,　Springer　Nature　Switzerland　AG.