Hi Kishor,

That's a very difficult problem, and I'm not sure ML can be a useful tool. The way I see it, this kind of events are better handled from a risk management framework.

As you pointed out, the model can predict the events because they happened before, though rarely.

Instead of forecasting, you could look into an anomaly detection method. For example, predict an anomaly if the predicted value deviates significantly from the mean or some other reference value.

Hope this helps