The　lack　of　historical　sales　data　and　word-of-mouth　information　in　the　film　preparation　period,　the　few　available　variables　and　the　uncertainty　in　the　prediction　process　lead　to　the　difficulty　in　predicting　the　total　box　office　demand　of　films.　To　solve　this　problem,　this　paper　constructed　and　verified　the　prediction　method　of　interval　reliability　demand　in　the　film　preparation　period,　which　combined　XGBoost　algorithm　and　D-S　evidence　theory.　Firstly,　the　total　box　office　interval　was　effectively　divided　according　to　the　sample　data　of　the　training　set,　and　XGBoost　was　used　to　complete　the　calculation　of　the　reliability　function　value　of　the　evidence　variables.　Then,　the　D-S　evidence　theory　was　used　for　information　fusion　to　obtain　the　results　of　box　office　interval　reliability　fusion.　Finally,　the　box　office　attribution　was　judged　by　the　interval　reliability,　so　as　to　realize　the　interval　reliability　demand　prediction　in　the　preparatory　period.　The　validity　of　the　proposed　method　was　verified　by　selecting　the　data　of　Chinese　films　from　2017　to　2019,　and　it　was　compared　with　the　classical　predictive　classification　algorithm.　The　results　showed　that　the　method　has　higher　prediction　accuracy　and　better　generalization　ability.　©　Journal　of　Physics:　Conference　Series　2021.