Evaluating the Reability and Efficiency of Hierarchical versus Non-Hierarchical Classification Methods
Conference
65th ISI World Statistics Congress 2025
Format: CPS Abstract - WSC 2025
Keywords: binary, classification, misclassification, process_efficiency, simulation
Session: CPS 7 - Epidemiological Modelling
Monday 6 October 4 p.m. - 5 p.m. (Europe/Amsterdam)
Abstract
The use of pooled classification methods can significantly save resources, but their reliability tends to decline with larger group sizes. Therefore, it is crucial to consider standard reliability measures, particularly sensitivity, which is most affected in these processes.
This study employs R-based simulations to identify scenarios where pooled classification methods are most appropriate, focusing on the trade-offs between efficiency (the number of tests performed) and reliability. Various pooled classification methodologies are compared, including hierarchical methods with varying stages and retesting configurations (such as the number of subgroups, their sizes, and their members at each stage), and non-hierarchical methods with and without a master pool.
The simulations apply both qualitative tests (with pooled sensitivity equal to individual sensitivity) and quantitative tests (considering the dilution effect). Different distributions for the discriminant substance (e.g., Gaussian, Weibull, Pareto, Lévy) were evaluated, along with different measures for the quality of individual tests, various prevalence rates (p in {0.0001, 0.001, 0.005, 0.01, 0.05, 0.1}), and group sizes.
Our results highlight the balance between efficiency and reliability, particularly sensitivity. The most suitable solution may not be the most efficient but rather an intermediate approach that provides substantial savings while minimizing the loss of reliability.