DFG funds project on anomaly detection in psychological data
The German Research Foundation (DFG) has approved funding for a new research project entitled "Anomaly Detection for Data with Missingness and Hierarchical Structure" led by Philipp Doebler. The project, funded through a DFG Individual Research Grant, focuses on advancing statistical and machine learning methods for detecting unusual patterns in educational and psychological data. The work will be conducted in cooperation with researchers at Meta Inc., London, UK, and the Hector Research Institute of Education Sciences and Psychology at the University of Tübingen, Germany.
The project addresses an emerging approach known as anomaly detection (AD). While traditional statistical analyses in the educational sciences usually focus on average effects or correlations across groups, anomaly detection aims to identify unusual individual observations. Such anomalies may arise from measurement artifacts, data quality issues, or rare but meaningful behavioral patterns. Detecting them can be valuable in many contexts—for example, identifying early signs of clinical relapse in psychology or recognizing when study participants disengage from a task.
Despite its potential, anomaly detection remains underused in educational and psychological research. Many modern AD methods originate from technical fields and often assume complete data, whereas psychological datasets frequently contain missing values and complex hierarchical structures such as repeated measurements within individuals. The studied approaches build on shallow ensemble-based anomaly detection methods. Although such methods often perform well in practice, their statistical properties are still not fully understood. By connecting advances in machine learning with the practical research needs, the project aims to enable more robust, interpretable, and widely applicable anomaly detection methods for educational and psychological data.
Contact person: Prof. Dr. Philipp Doebler, doebler@statistik.tu-dortmund.de