An Outlier Detection Strategy for High‐Dimensional Data in Industrial Processes

This paper proposes a new approach to explore how biplots can be used to detect outliers and uncover emerging patterns in high‐dimensional data. The strategy begins by generating a biplot from typical in‐control data, offering a two‐dimensional visualization of multivariate relationships. Test samples are then projected onto this biplot, with their positions compared to the reference data cloud. Samples that fall outside this cloud are flagged as outliers, indicating deviations from expected behavior. This procedure simplifies outlier detection and the analysis of underlying causes without relying on assumptions about data distribution. A real‐world example from a mineral turpentine production process, involving 10 control variables, demonstrates the effectiveness of this approach. It was found to be intuitive, helped identify hidden outliers in the early stages, provided valuable insights for process improvement, and showed potential for application across various fields.