This has the same issue as many other types of event warning systems based on noisy, incomplete data.

The latency of constructing a semi-reliable warning signal from the data sources described significantly exceeds the latency of event onset. You can modify the algorithms to reduce latency but then the false positive rate skyrockets. Not what you want for an "apocalypse" early warning system.

To mitigate this you need more data from more diverse sources and lower latency feeds.