Abstract—Althoughtherehasbeensubstantialresearchinsoftwareanalyticsforeffortestimationintraditionalsoftwareprojects,littleworkhasbeendoneforestimationinagileprojects,especiallyestimatinguserstoriesorissues.Storypointsarethemostco妹妹onunitofmeasureusedforestimatingtheeffortinvolvedinimplementingauserstoryorresolvinganissue.Inthispaper,weofferforthefirsttimeacomprehensivedatasetforstorypoints-basedestimationthatcontains23,313issuesfrom16opensourceprojects.Wealsoproposeapredictionmodelforestimatingstorypointsbasedonanovelcombinationoftwopowerfuldeeplearningarchitectures:longshort-ter妹妹emoryandrecurrenthighwaynetwork.Ourpredictionsystemisendto-endtrainablefromrawinputdatatopredictionoutcomeswithoutanymanualfeatureengineering.Anempiricalevaluationdemonstratesthatourapproachconsistentlyoutperformsthreeco妹妹oneffortestimationbaselinesandtwoalternativesinbothMeanAbsoluteErrorandtheStandardizedAccuracy.
1