Pursuing efficient bifunctional oxygen reduction reaction/ oxygen evolution reaction (OER/ORR) electrocatalysts is crucial for realizing sustainable and renewable clean energy, yet it remains a formidable challenge. In this study, we employ density functional theory and machine learning (ML) predictions to systematically investigate the bifunctional OER/ORR activity of 3d transition metal (TM) atom-doped g-C₃N₃. Our findings reveal that N-rich and C-poor growth conditions can facilitate 3d TM doping into g-C₃N₃. The Ni occupation N in 0 (Ni^×_N@C₅₄N₅₃, including 54 C and 53 N atoms), Ni occupation N in +1 (Ni^•_N@C₅₄N₅₃) and Ni interstitial sites in +1 charge state (Ni^•_int@C₅₄N₅₄, including 54 C and 54 N atoms) meet the requirements of bifunctional OER/ORR electrocatalysts, which because they (1) have lower formation energies, and (2) lower overpotentials (smaller than 0.47V), (3) can maintain stability under 300K and (4) their band gaps are decreased, which leads to facilitating electron transfer. ML predictions identify the 3d band center (ε_d) and the number of 3d electrons (N_e) as the most effective descriptors for overpotentials of OER and ORR, respectively. This work can pave the way for understanding the origin of bifunctional OER/ORR activity of 3d-TM doped g-C₃N₃ and benefit the rational design of novel electrocatalysts for other catalytic reactions by considering the charge states.