Due　to　the　interference　of　various　factors,　the　experimental　data　are　filled　with　a　large　number　of　untrustworthy　data,　which　brings　great　difficulties　to　the　prediction　of　PM2.5.　In　view　of　this　problem,　this　paper　establishes　a　data　credibility　measurement　model　to　analyze　the　credibility　of　the　data.　Firstly,　the　credibility　between　data　sources　is　calculated　based　on　the　similarity　between　data　sources.　Secondly,　the　direct　credibility　of　the　data　source,　the　recommended　credibility　of　the　data　source　and　the　penalty　value　of　the　data　source　are　used　to　calculate　the　comprehensive　credibility　of　the　data　source.　Finally,　the　credibility　of　the　data　is　calculated　based　on　the　opposite　event　in　which　all　data　sources　provide　erroneous　data.　The　method　is　applied　to　the　prediction　of　PM2.5　concentration,　and　the　credibility　of　air　quality　data　is　analyzed.　The　experimental　results　show　that　the　model　can　calculate　the　credibility　of　the　data　sensitively　and　effectively,　which　increases　the　accuracy　of　PM2.5　prediction　results　and　provides　a　solution　for　further　study　on　data　credibility　evaluation　methods.　©　2019　IEEE.