Umeå universitets logga

Financial markets generate vast amounts of data, making it increasingly challenging for analysts to identify meaningful patterns and insights. This thesis explores clustering techniques for categorizing trader firms within the Nordic equity markets, incorporating a human-in-the-loop approach to enhance interpretability. Using order book data from Nasdaq Nordic, trader behavior is analyzed, and k-means, hierarchical, and spectral clustering algorithms are applied to classify firms based on their trading patterns.

Due to the lack of ground truth, clustering effectiveness is assessed through intrinsic evaluation metrics and expert validation, revealing a discrepancy between intrinsic measures and expert assessments. The results underscore the importance of human-in-the-loop approach and demonstrate the potential of unsupervised learning to support market analysis. This research bridges machine learning methodologies with domain expertise and provides a foundation for future clustering-based analytics in financial markets.