Constructor University wins Big Data Challenge with a new fraud model!

Constructor University wins Big Data Challenge with a new fraud model!

Bremen, Deutschland - A team of Constructor University won in the professional track. The award was awarded for an innovative, transparent and regular statistical model for the recognition of financial fraud. This model is not only characterized by its precision, but also exceeds conventional methods of machine learning in accuracy.

The competition, which is organized annually by the Cognitive Systems Lab (CSL) of the University of Bremen, focused on the fraud recognition this year. Participating teams, including leading AI companies in the region, faced the challenge of developing effective approaches to identify fraudulent activities. The team that competed under the name Spiderbobs and under the direction of Dr. Johannes Falk, postdoctoral in the Computational Systems Biology group, operated, consisted of Eda Cakir and Dr. Ali Salehzadeh-Yazdi.

innovative methods for fraud detection

For solving the problem, various common models, including LStM neuronal networks, were tested. In the end, however, the team developed a rule-based model that is characterized by an F1 score of 0.9992, which is considered a central accuracy indicator in fraud detection. The teams aimed at creating transparent analyzes that are able to compete with the often non-transparent black box methods of the AI ​​or even exceed them.

The developed model requires a deep understanding of data mechanisms and probabilistic modeling. It is based on clearly defined rules and probabilistic considerations and analyzes money flows and network structures. This methodology enables the identification of non -fraudulent accounts and the creation of behavioral profiles for legitimate account movements. The two -stage strategy applied includes the comparison with the behavioral profile and the review of fraud probability.

data analysis and fraud prevention

This days, terms such as big data, predictive analytics and machine learning are becoming increasingly important. They revolutionize business models and can also be used for fraud prevention. According to Risknet are the use of statistical methods for the analysis and visualization of large data records at the center of this development. The aim of the data analysis is not only the creation of personality profiles, but also the implementation of real-time predictions.

Das Benford's law, discovered by Simon Newcomb in 1881, is an example of the use of statistical analyzes for the detection of fraud. However, more complex methods, including machine learning, significantly increase the prognosis bag of risk models. The PDCA cycle, which enables structured data analyzes, and the KDD process (Knowledge Data Discovery) are central elements for effective wife management.

With the success of the team from Constructor University, once again, the importance of interdisciplinary research and scientific excellence in the field of data analysis and fighting fraud is underlined. In view of the progressive digitization, companies have to develop a new understanding of data analyzes in order to optimally use the opportunities of today. Future risk managers will increasingly act as data managers or scientists and develop robust data analysis models.