When a volcano becomes restless, it is challenging to assess whether it will lead to an actual eruption and determine the timing of the eruption onset. A magmatic intrusion starting at depth can (a) remain at depth, (b) stall just before reaching the surface, (c) erupt in sluggish and viscous extrusion, or (d) erupt rapidly or explosively (Moran et al., 2011). The process of magma migration involves interactions with the surrounding country rock, cooling magma bodies from previous eruptions, and (or) hydrothermal system (Moran et al., 2008). These interactions generate natural phenomena such as earthquakes, deformation, temperature changes, and gas emissions. These phenomena can be observed by geophysical and geochemical measurements (Moran et al., 2008) and integrated with the history of past eruptions in a framework of eruption forecasting (Whitehead & Bebbington, 2021).From a seismic point of view, eruptions can show precursors such as accelerating or decelerating earthquake rates. To assess this, monitoring institutes conventionally use methods to tabulate daily event counts (McNutt, 1996) and calculate the average amplitude for a certain window length (Endo & Murray, 1991). The Failure Forecast Method estimates the onset time of eruption by using the rate and the acceleration of seismic precursors associated with the rock failure caused by magma propagation (Boué et al., 2015). However, this method cannot deal with complex precursory signals, for example, that exhibits fluctuations or deceleration (Boué et al., 2015). Furthermore, due to the uncertainty of the eruption forecast and numerous false alarms (Bell et al., 2013), this method is not recommended to be stand-alone (Whitehead & Bebbington, 2021). Dempsey et al. (2020) tested a real-time Machine Learning framework to detect eruption precursors of five major eruptions at Whakaari