In　order　to　improve　the　traffic　safety　problems　caused　by　electric　bicycles,　this　paper　studies　the　relationship　between　reverse　riding　risk　behavior　and　its　influencing　factors.　Based　on　the　GPS　trajectory　data　of　shared　E-Bike　in　Furong　district,　Changsha　City,　the　accurate　identification　of　reverse　riding　behavior　is　realized.　The　machine　learning　model　CatBoost　and　interpretable　machine　learning　framework　SHAP　are　used　to　extract　and　analyze　the　influencing　factors　of　reverse　riding　behavior　from　the　aspects　of　road　conditions,　traffic　conditions,　land　use　attributes,　etc.　The　results　show　that:　①　CatBoost　model　can　effectively　predict　the　reverse　riding　frequency　of　road　sections　and　extract　the　important　influencing　factors　of　reverse　riding　behavior,　mainly　including　travel　time,　public　transport　facilities,　land　use　attributes,　road　conditions　and　traffic　conditions.　②　In　terms　of　travel　time,　reverse　riding　is　more　likely　to　occur　on　weekdays　and　morning　&　evening　peak　hours;　In　terms　of　public　transport　facilities　and　land　use　attributes,　the　number　of　bus　stops　and　subway　station　exits,　and　the　number　of　restaurants,　companies,　shopping　and　other　facilities　around　the　roads　present　a　non-linear　influence　relationship　to　the　reverse　riding　frequency.　In　a　certain　range,　the　number　of　facilities　has　a　positive　effect　on　the　reverse　riding　behavior.　In　terms　of　road　conditions,　reverse　riding　is　less　likely　to　happen　with　road　crossing　intervals　of　50~400　m.　But　reverse　riding　is　more　likely　to　happen　when　there　are　no　physical　separation　facilities　in　the　bicycle　lane　or　with　road　crossing　intervals　of　400~600　m.　And　the　effect　is　unstable　when　the　intervals　is　wider　than　600　m.　In　terms　of　bicycle　lane,　the　reverse　riding　probability　with　guardrail　separation　is　lower,　while　the　probability　with　green　belt　separation　is　higher.　In　terms　of　traffic　conditions,　when　the　riding　speed　and　acceleration　are　too　low　or　too　high,　it　is　negatively　related　to　the　reverse　riding　behavior.　When　the　riding　speed　is　between　6~16　km•h-1　and　the　acceleration　is　between　0.3~1.0　m•s-2,　it　is　positively　related　to　the　reverse　riding　behavior.　This　research　can　provide　an　effective　technical　support　for　the　identification　of　shared　E-Bike　risk　riding　behavior,　as　well　as　for　the　management　of　non-motor　vehicle　traffic　safety.　©　2021,　Editorial　Department　of　China　Journal　of　Highway　and　Transport.　All　right　reserved.