Data science in customer services
The model is built and based on transactional data and domain experts’ (i.e. law enforcement officials at the frontline) knowledge. It entails detecting (non)obvious patterns in large amounts of municipal data by using pattern recognition. Underlying to pattern recognition is the machine learning algorithm ‘decision three’. As such, decision trees are compatible with human-driven processes. In this way, it is possible to detect anomalies that might result from people committing benefit fraud.
Instead of reviewing data after the moment of transaction, it is feasible to examine new transactions in real-time. With the approach called 'feature modelling' relevant information about applicants is used to make this possible. This can lead to a more efficient way of detecting likely fraudsters by the law enforcement department. In this way, transactional data together with domain experts’ knowledge as part of an iterative model design process can create a proper system to detect fraud. We can proactively identify risks.
Since the model seemed to work for the benefits services, Groningen is currently applying similar models for other services, for instance waste-bins (people tend to put waste on the street instead of in the bins) and for improving the quality of the citizens register. The approach is roughly the same albeit the subject is different. Scanning data and applying expert knowledge results in better quality of the city’s ledger and of the detection of patterns in the waste depositions.