The Sun sends out energetic particles that interact with Earth's magnetic field, sometimes causing sudden shifts in the magnetic field. These changes can damage infrastructure, so it's crucial to predict when and where they will happen to reduce their impact. However, these disturbances occur over large areas and in specific locations. To improve forecasting, we use machine learning (ML), an advanced method for identifying patterns in data. The challenge with ML is that it can be difficult to understand exactly how it makes predictions. Our study trains ML models using solar wind data and magnetic field measurements. We aim to predict two key space weather events: sudden changes in Earth's magnetic field and differences in magnetic activity across regions. Our models make forecasts for a 60-minute period, starting 30 minutes into the future, and they perform well across different locations. We find that commonly known space weather drivers, like certain properties of the solar wind, are important for our models. However, we also identify other factors that are not often studied but play a significant role. Additionally, we observe that the way Earth's magnetic field changes with time on a regional scale differs from previous findings based on individual locations.