Learning　engagement　is　regarded　as　an　important　metric　for　assessing　E-learning　since　it　is　related　to　the　quality　of　online　education　and　student　performance.　Previous　studies　have　identified　a　correlation　between　heart　rate　variability　(HRV)　and　learning　engagement.　However,　a　comprehensive　and　systematic　investigation　of　the　underlying　mechanisms　remains　lacking.　In　this　study,　a　GRU-TCN　model　was　developed　to　classify　learner　engagement　levels　in　E-learning　based　on　photoplethysmography　(PPG).　PPG　signals　were　collected　from　40　participants　as　they　learned　the　video　course　under　task　conditions　of　no-interference,　music　&　bullet　chat　stimuli,　and　mental　arithmetic.　Twenty-eight　HR/HRV　metrics　(including　time-domain,　frequency-domain,　and　Non-linear　metrics)　were　examined,　among　which　10　metrics　were　sensitive　to　changes　in　learner　engagement.　To　adjust　for　individual　differences　in　heart　response　in　learning　tasks　and　to　enhance　the　data,　a　three-step　processing　procedure　was　conducted　to　evaluate　the　PPG　signals　of　each　participant:　(1)　defined　three　levels　of　learning　engagement:　full　attention,　external,　and　internal　divided　attention,　(2)　normalized　HRV　metrics　for　different　engagement　levels　by　using　the　average　metric　of　participants＇　baseline　states,　(3)　Enhanced　data　with　sliding　time　windows.　The　GRU-TCN　classification　model　achieved　F1-scores　of　95%　in　the　training　set　and　85%　in　the　testing　set,　respectively,　enabling　effective　automatic　real-time　tracking　of　learner　engagement　in　Elearning.