Understanding Distribution Evolution in High Dimensional Data Streams

This research explores time-varying multidimensional numeric data, focusing on how machine learning models can remain robust when data distributions shift over time (concept drift). The thesis presents methods to measure and manage this drift in high-dimensional settings, ensuring better model accuracy. It also introduces a summary-based forecasting approach that predicts entire distributions, rather than individual points, to guide long-term planning. These insights offer practical tools for handling complex, evolving datasets and contribute to more reliable decision-making in real-world applications.