As　the　most　widely　distributed　pollutant　in　the　earth＇s　atmosphere,　CO　has　received　extensive　attention　and　research　worldwide.　The　CO　concentration　is　now　detectable　by　remote　and　sensor　methods.　In　this　paper,　a　linear　regression-based　calibration　method　for　CO　sensors　is　proposed,　in　which　the　compensation　of　the　gas　sensor　drift　is　accomplished　by　iteratively　searching　the　training　data　set　containing　the　gas　information　to　derive　the　best　composition　parameters　that　minimize　the　error　value.　In　this　experiment,　the　root　means　square　error　between　the　model　results　of　the　gradient　descent　algorithm　and　the　actual　values　was　0.235.　Subsequently,　the　model　accuracy　was　improved　by　further　fitting　the　model　results　to　the　actual　values　through　a　locally　weighted　linear　regression　algorithm　by　increasing　the　weights　of　the　data　points,　which　reduced　the　resulting　RMSE　by　31%　to　0.162.　The　experimental　results　achieved　reliable　results　and　small　error　values..　The　sensor　can　be　used　in　various　fields　for　the　detection　and　prediction　of　CO　content　to　ensure　the　safety　of　people＇s　lives　and　also　to　provide　more　possibilities　for　the　compensation　of　gas　drift　response　studies.　©　2022　IEEE.