Adaptive Multi-Modal Dimensionality Reduction Framework for High-Dimensional Streaming Data Clustering

High-dimensional data clustering faces three critical challenges: increased computational complexity, curse of dimensionality, and degraded clustering performance. We evaluate preprocessing techniques that effectively address these dimensionality-related hurdles while preserving the intrinsic data structure. Our systematic comparison of dimensionality reduction methods reveals that a systematic trade-off between Variational Autoencoders, First Neighbour Relations and time-tested technique like Principal Component Analysis significantly reduces computational overhead while maintaining clustering quality. The findings provide a robust preprocessing framework for high-dimensional clustering tasks.