Conclusively linking local, episodic enhancements in greenhouse gas concentrations to a specific emission source can be challenging, particularly when faced with multiple proximal sources of emissions and variable meteorology, and in the absence of co-emitted tracer gases. This study demonstrates and evaluates the efficacy of using machine-learning tools to detect episodic emissions of methane (CH4) from a shale gas extraction facility in Lancashire (United Kingdom). Two machine-learning tools (rmweather and Prophet) were trained using a two-year climatological baseline dataset collected prior to gas extraction operations at the facility. The baseline dataset consisted of high-precision trace gas concentrations and meteorological data, sampled at 1 Hz continuously between 2016 and 2019. The models showed good overall predictive capacity for baseline CH4 concentrations, with R2 values of 0.85 and 0.76 under optimised training conditions for rmweather and Prophet, respectively. CH4 concentrations were then forecast for an 18-month period from the onset of operations at the shale gas facility (in 2018). Forecast values were compared with true measurements to detect anomalous deviations that may indicate the presence of new emission events associated with the operational facility. Both models successfully detected two periods in which CH4 emissions were known to have occurred (December 2018 and January 2019) via anomalous deviations between modelled and measured concentrations. This work demonstrates the application of machine-learning models for the detection of CH4 emission events from newly built industrial sources, when used in combination with real-time atmospheric monitoring and a baseline dataset collected prior to installation.