RAG has been producing hard coal by underground mining in Germany until the end of 2018 and continues to be responsible for the resulting consequences in North Rhine-Westphalia and Saarland. For almost 200 years, thousands of mine shafts and other cavities have been developed underground, many of them at a time when computers and 3D models were still unknown.
By performing continuous monitoring of ground anomalies RAG is able to identify potentially impacted areas early. (One very successful algorithm actually resulted from an previous Data Hub Challenge “spatial pattern recognition“.) But despite all prevention efforts property damages caused by (former) mining activities are still common in the region. RAG is liable for many of those damages and compensates affected owners. Therefore, a strong understanding of expected claims and compensation costs is critical to RAG business. In addition to decades of individual expertise, experts use already established prediction models based on size and population density of the shutdown area.
The goal of this use case is to use historical data of property damages to better predict future compensation costs. It is believed that the number and value of damages claimed in all shutdown areas is following a similar “curve” over time. Therefore, historical claim data from areas already shutdown for a long period should allow us to predict claim data of just recently stopped mining areas. We expect a more precise prediction than current models and an earlier recognition of potential overspending of provisions.
An additional way to improve prediction quality would be to consider the locality of claims. For example, if similar claims by 3 of 5 houses in a street are received, the probability of claims by the other 2 house owners is relatively high.
Data of multiple thousands of claims per year from up to 43 shutdown areas is available. Especially, claim data since 2000 can be considered complete and high-quality (date, value, kind of damage, kind of building, location).
The project can be divided into three milestones: