Seismic hazard is the hardest detectable and predictable of natural hazards and in this respect it is comparable to an earthquake. More and more advanced seismic and seismoacoustic monitoring systems allow a better understanding rock mass processes and definition of seismic hazard prediction methods ... It is essential to search for new opportunities of better hazard prediction, also using machine learning methods. (Sikora M. et al, 2010)

This challenge is to make use of this now rather outdated but still well-published and well-known reference dataset. Feel free to suggest any more recent open publications that we should add to DataHub.

Simon Wenkel's exploration of this dataset includes performance comparison of various algorithms, as well as these excellent follow-up questions:

• Where are the geophones located? Is it possible to do TRM (time reverse modeling) of stress wave propagation? With this dataset we simply know that if a rock burst occurs in a longwall but not where.
• What to do with the results? We may end up knowing if a rock burst is likely to occur or not. Can we use the recorded data to estimate locations and therefore controll stress release before the burst?
• Can we extract another set of useful information from the raw data from the geophones (what we got here is heavily processed)?

Shravan Kuchkula and Nabanita Roy's articles (Part 1, Part 2) do a great job of visualizing the statistical distributions and suggesting prediction model approaches. For a more formal description, see Enes Çelık et al 2014.

Fig. 4 Deep learning workflows for earthquake prediction, Arnaud Mignan and Marco Broccardo 2019

📦 File: datapackage.json

📦 Data: HCLNOGAlevels_4-5.csv