Kilauea is the main active volcano on the island of Hawaiʻi and has been nearly constantly active since 1983, posing a hazard to life and infrastructure. Current volcanic prediction methods, both numerical and machine learning–based, remain limited and are not yet widely implemented. This study attempts to forecast the probability of eruption at Kilauea using remote sensing data from Sentinel-1A/B C-band synthetic aperture radar (SAR). These data are processed to produce interferograms (InSAR images), from which ground deformation estimates are used as input to a Convolutional Long Short-Term Memory (ConvLSTM) model that predicts the likelihood of future eruptions. The ConvLSTM approach is well suited for spatiotemporal datasets, as it captures spatial features and time-dependent relationships simultaneously. The model is trained on a spatiotemporal dataset of real and synthetic deformation maps derived from InSAR over the Kilauea caldera and the broader island of Hawaiʻi. Built using TensorFlow’s ConvLSTM2D library, the model is designed to predict eruption probability at a temporal resolution of two weeks or less, with the broader goal of developing a transferable framework that can be applied to other volcanoes and integrated into hazard assessment efforts.