Understanding earthquake source processes is crucial for hazard assessment, particularly in tectonic environments like northeastern British Columbia, where hydraulic fracturing (HF) has significantly increased injection-induced earthquakes (IIEs). This study integrates advanced observational techniques, machine-learning algorithms, and statistical seismology to delineate the physical and operational mechanisms driving IIEs. Initially, we constructed enhanced seismic catalogs for the Montney Play, utilising AI-driven phase picking to lower detection thresholds and improve location accuracy. Utilizing eXtreme Gradient Boosting interpreted through Model-Agnostic Explanations, we evaluated various factors contributing to the seismogenic potential and productivity of individual HF pads. The results reveal that cumulative injected volume and proximity to structural features, specifically the Fort St. John Graben (FSJG), are the primary controls on seismic occurrence. Furthermore, our analysis identifies a critical stratigraphic control, with over 90% of IIEs in the Montney Play associated with HF targeting the Lower-Middle Montney (LMM) formation. We employed a seismogenic index approach to quantify regional susceptibility and forecast maximum IIE magnitudes. Retrospective validation demonstrates that our models effectively capture the spatial distribution of IIEs, with forecasted maximum magnitudes averaging about 15% higher than observed values, reflecting a conservative probabilistic approach. Furthermore, our forecasting of expected event numbers aligns most consistently with observed seismicity at larger magnitude thresholds (M≥3). The contrast between volume-triggered nucleation within the FSJG versus stratigraphy-driven event frequency in the LMM suggests that the mechanisms governing seismogenic potential and productivity are distinct. These findings establish a comprehensive framework for seismic hazard characterisation, supporting effective mitigation strategies like refined traffic-light protocols.