Investigating data quality issues in Data Streams by evaluating the effects of Label and Feature Noise

Data quality is a critical determinant of machine learning performance, particularly in real-time applications where models continuously adapt to incoming information. While the adverse effects of class (label) and attribute (feature) noise are well-documented in static learning settings, their comparative influence in streaming environments remains underexplored. This study conducts a comprehensive empirical analysis to assess how these two types of noise affect model performance under streaming conditions. Artificial noise is systematically introduced into incoming data batches, and model responses are evaluated using standard metrics. The results consistently show that label noise has a more substantial and destabilizing effect on predictive accuracy than feature noise in dynamic data scenarios. These findings underscore the need for effective noise mitigation techniques in real-time machine learning pipelines to ensure robust decision-making in the presence of unreliable or inconsistent.