Clustering of workers with similar learning profiles based on Principal Component Analysis

The manufacturing of customized products, also called mass customization, implies increasing choice menu and reduced size of production batches. Tasks that depend on human ability are especially affected in this context, because workers need to adapt to the characteristics of new models. This adaptation process occurs differently among workers, justifying the development of systematics aiming at clustering workers with similar learning behaviors. This paper proposes a method of grouping workers according to their learning profiles, integrating learning curves (LC) and cluster analysis. To this end, performance data are collected and modeled through learning curves; parameters derived from the modeling quantify workers' adaptation to tasks. In the first scheme proposed in this study, the original data (parameters from LC modeling) are clustered using the K-Means, and clustering quality is assessed by the Silhouette Index (SI). In the second scheme, Principal Component Analysis (PCA) is applied to the original data and the latent variables (scores) are used as clustering variables using the K-Means. Clustering using the scores yielded SI=0.968, while grouping based on the original variables led to SI=0.392. A simulation study was performed to corroborate the effectiveness of the proposed method, which proved to be robust when affected by different levels of noise, correlation and different proportions of variables to observations.

Learning curves; Clustering; Principal component analysis