Improving High Dimensional Data Streams Classification

Traditional data mining approaches can no longer meet the analytical needs of modern Big Data, as many of today's applications generate an infinite amount of dynamic data known as “data streams”, in real time and at rapid rates. In response, data stream mining has emerged as a new paradigm for processing and deriving real-time insights from such data. Data stream mining encompasses a variety of techniques, with stream classification being the focus of much research. Accordingly, this research aims to refine one of the key methods employed in online classification. This study explores the application of Adaptive Random Forests (ARF) to data stream classification, with a focus on the Compressed-Sensing ARF (CS-ARF) variant, which has been specifically designed to tackle the challenges of high-dimensional data streams classification. Based on the CS-ARF implementation, our research aims to further enhance its performance. Experimental results demonstrate promising outcomes, particularly in terms of improved classification accuracy.